Projects

Abstract

The ability to study tissue microstructure in vivo and completely noninvasively using magnetic resonance imaging (MRI) has the potential to radically change how we detect, monitor, and treat diseases, in particular the many neurodegenerative diseases that affect our world's aging population. Unfortunately, the MRI signal is a very indirect measure of microstructure, and the variety of contributing factors complicates a one-to-one association between the MRI measurements and the biological substrate. As a result, microstructural mapping is still a poorly understood and challenging inverse problem that often yields inconsistent and contradictory outcomes. In ADAMI, I will take the next leap in microstructure imaging by approaching the problem in a completely data-driven fashion as opposed to the state of-the-art that is model-driven. This paradigm shift will enable me to turn the MRI scanner into a powerful in vivo microscope that can provide reliable information about tissue microstructure that closely matches the underlying cellular composition. Through these innovations, ADAMI will advance the field of medical imaging by introducing a groundbreaking data-driven approach to microstructure imaging which will significantly impact the understanding, diagnosis, and monitoring of brain diseases and beyond.

Researcher(s)

• Promoter: Jeurissen Ben

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

Diffusion magnetic resonance imaging (dMRI) is a powerful MRI modality for in vivo imaging of biological tissues and an invaluable tool for pathophysiological understanding several neurological disorders such as stroke, brain tumors and cerebral infections. Unfortunately, the traditional dMRI workflow (consisting of data acquisition, image reconstruction, postprocessing, and parameter estimation) is time consuming, both in terms of acquisition and processing time. Moreover, the current multi-step, unidirectional dMRI workflow leads to error propagation. This hinders implementation in routine clinical practice. In this project, we will radically change the dMRI workflow by developing UNIPHY, a novel, unified framework that substantially improves on the quality and speed of the dMRI workflow. First, we introduce a new acquisition strategy with more flexibility for optimal experiment design, with a potential 2-fold scan time reduction. Second, we will include motion and distortion estimation directly into the dMRI parameter estimation, thereby avoiding error propagation. Finally, we will introduce state-of-the-art physics-informed deep learning to accelerate processing by at least 10-fold.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

FleXCT is an advanced X-ray system, unique in its kind as it has 10 degrees of freedom. It is used to mimic various industrial X-ray environments. The maintenance cost of the system is about 20k euro/year. This funding covers 25% of it.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main goal of this project is to develop a new data analysis approach for quantitative material characterization from spectral reflectance. If successful, the proposed approach will enable accurate estimation of material properties, in a non-destructive and non-contact manner, unlike many existing laboratory-based measurement techniques. Moreover, it is employable in situ, for the inspection of large infrastructures or quality control in production. The proposed approach combines a geometric description in which material properties are represented on curves or low-dimensional surfaces, and machine learning to relate these representations to actual estimates of material properties. The representation is made invariant to spectral variability, to make it applicable under variable environmental and acquisition conditions and in cross-sensor situations. The proposed methodology will be developed for a large group of material characterization tasks. More specifically, methods are developed for estimating the material composition (i.e., the mass fractions of the material components), detecting target material components, and estimating water content. Furthermore, three use cases of the developed methodologies will be elaborated, related to corrosion monitoring and characterization of building materials.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Bnoulkacem Zakaria

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

A material can be uniquely characterized by its optical reflectance spectrum. Hyperspectral cameras disperse the reflected sunlight into hundreds of consecutive small wavelength bands in the visible and near-infrared (VNIR, 400-1000 nm) and the shortwave infrared (SWIR, 1000-2500 nm) regions. The main objective of this project is to develop advanced innovative spectral analysis methods that relate optical reflectance to material properties. I will focus on 3 particular material properties, for which a framework will be developed, validated, and applied on specific case studies: 1. mineral composition estimation, with a case study in geological mining. 2. plant leaf biochemical parameter estimation, with a case study in multi-scale forest ecological functioning 3. water content estimation, with a case study in climate change effects on built heritage

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Koirala Bikram

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

3D printing, and additive manufacturing in particular, has in the last couple of years transformed from a prototyping to a more mature manufacturing technology. The driving force for this transformation was the need for custom built, lightweight parts and a more efficient use of raw materials in order to reduce waste. The current trend is towards increasingly complex parts, consisting of multiple materials to enhance their physical and/or mechanical properties. The complexity of the parts brings with it a need for specialized inspection and quality control methods. In this project, the complementarity between different imaging technologies such as 3D x-ray imaging, thermal imaging and terahertz imaging will be explored within the application field of 3D inspection and metrology of 3D printed parts.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Billen Pieter
• Co-promoter: Sijbers Jan
• Co-promoter: Vanlanduit Steve
• Co-promoter: Van Passel Steven

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The Pacifix II project is a continuation of the results of the Pacifix I project. During the Pacifix I project we developed a framework to segment and reduce (put back together) the fragments of a distal radius bone fracture. In the Pacifix II project we will develop software tools to automatically prepare the surgery. The software will determine which surgical plate is the best suited for the patient and his fracture by virtually fitting the plate on the reduced bone. It will generate a pre-operative plan with an optimized position of the plate. Furthermore, the position and length of the screws will be optimized. To reduce the time in the operation room, surgical guides that are a physical template of the plate, screws and bone will be developed. Therefore, there is only one way to fixate the plate and screws to the bone.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Vanhees Matthias

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Conventional X-ray imaging based on attenuation contrast is widely known through e.g. medical chest radiographs. By acquiring thousands of X-ray radiographs, a 3D representation of the microstructure of an object can be reconstructed, which has many applications in industrial inspection and metrology of manufactured materials. X-ray imaging offers two more contrast types: phase contrast (due to refraction at interfaces) and dark-field contrast (ultra-small angle scattering). The latter two effects can only be observed with specialized grating hardware, such as in the edge-illumination (EI) technique. Phase contrast can be up to 1000 times stronger than attenuation contrast for 'soft' materials, such as polymers. Unfortunately, EI requires at least three measurements to separate the different contrasts, leading to long processing times. This project aims at exploiting phase contrast properties for efficient inspection and metrology of manufactured objects. To limit the number of measurements, algorithms equipped with prior knowledge in the form of 3D mesh models of investigated samples will be developed. Few-view inspection and metrology techniques will then be developed in which measured phase contrast radiographs are compared to simulated radiographs from the reference CAD projections. To enable metrology, adaptability of the surface mesh to acquired radiographs will be implemented. The methods will be validated on manufactured objects containing plastics and metals.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan
• Fellow: Merckx Jannes

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

With technological progress, small size, low cost multispectral and hyperspectral cameras become available that capture the light in up to hundreds of consecutive small wavelength bands in the visible and near infrared (VNIR, 400-1000 nm) and the shortwave infrared (SWIR, 1000-2500 nm) regions. These cameras can be installed on unmanned aerial vehicles, agricultural equipment or conveyer belts, or used in laboratory environments and even be used as handheld devices. Numerous applications can be considered for industrial inspection and quality control. As water has very strong absorption power, in particular in the SWIR range (e.g., absorption peaks around 1400 and 1900 nm), the optical reflectance properties of water-bearing materials are dominated by water. This can be disadvantageous when characterizing materials, as the material reflectance characteristics are largely hidden because of the water absorption. On the other hand, this gives opportunities to focus on water-related properties of a material, e.g., its water content, or specific material parameters that can be related to water content, e.g., leaf physiological parameters such as the Equivalent Water Thickness. The goal of this project is to study the characterization of water-bearing materials from optical reflectance imaging and to estimate and spatially resolve the water content and other relevant water-related material parameters. The main novelty and challenge is the development of a hyperspectral image analysis framework that is • invariant to environmental and acquisition conditions, • generically applicable to a large group of materials. These improvements will allow to upscale from point-based laboratory applications on benchmark datasets to spatially resolved real world in situ applications. The developed framework will be validated on two specific use cases: moisture content estimation in soils, and plant leaf parameter estimation.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Jambhali Ketaki Vinay

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Additive manufacturing (AM) is driving a design and industrial revolution in sectors such as aerospace, energy, automotive, medical, tooling and consumer goods. By 2027, the AM market is expected to almost triple in size to over €30 billion. In AM, ~30% of parts are made layer-by-layer by fusing metal or polymer with a high-power laser. This laser-based AM (LBAM) includes the most common printing techniques: laser powder bed fusion (LPBF) for metals; selective laser sintering (SLS) of polymers. LBAM has limitations in both productivity and efficiency. Roughly 10% of prints must be scrapped due to various defects, resulting in costly post-build inspections as target applications are typically high-end and safety critical. Therefore, the clear need is to improve print quality and reduce costly scrap. Newer LBAM printers increase productivity by having multiple lasers fuse material. This introduces new defect formation mechanisms: misalignment between scan fields; thermal interactions between nearby lasers; interference from spatter particles; laser diffusion from nearby vapor plumes. All lasers need to be controlled in a coordinated way to avoid these defects by a higher level of control over local thermomechanical conditions. Also, the monitoring system needs to cope with higher data rates. Traditional off-axis monitoring systems require one camera per laser, and this solution does not scale well (prohibitive cost, installation complexity). An off-axis in-line monitoring system obviates the need for a dedicated sensor per laser, resulting in a scalable solution and making the monitoring system integratable to existing machines. No such system exists. The above needs and gaps in state-of-the-art (SOTA) translate to the following research themes tackled by MultipLICITY: 1) Fusing information from multiple sensor types to detect a wider range of defect producing conditions. This data fusion needs to operate in a real time control loop, thus requiring research on resource-constrained fusion and analysis algorithms; 2) Generic, multi-material type printer control and defect detection, requiring limited or no retraining. MultipLICITY aims to increase the quality, productivity, and efficiency of LBAM by expanding in-line monitoring and control to multiple defects, multiple materials, and multi-laser systems, at a competitive cost.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

With this project we aim to acquire fast terahertz detection hardware as part of a full-field imaging system. With the new imaging equipment, we will be able to acquire new experimental data that is vital for research in CT reconstruction algorithms and we will be able to extend tomographic reconstruction concepts from x-ray imaging towards the THz domain.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

X-ray computed tomography (XCT) is widely used in material sciences and industrial applications (e.g., non-destructive testing and inspection) for non-invasive imaging. While traditional XCT relies on absorption of X-rays to generate image contrast, phase contrast X-ray imaging allows to additionally measure the local scattering power in the sample as well as the phase shifts of the X-rays. Edge Illumination (EI) is a phase contrast imaging technique suitable for use with conventional, polychromatic X-ray sources, which has demonstrated its great potential for translation into real-world environments. Unfortunately, adoption of EI-XCT in industry is slow since it requires (up to 4 times) more X-ray data to be acquired, leading to substantially higher acquisition times compared to traditional XCT. In this project, I will develop acquisition and reconstruction methods for EI-XCT with scan times comparable to those of traditional XCT, while still providing the three complementary contrasts. Furthermore, the acquisition method is paired with a dedicated iterative reconstruction algorithm for increased quantification. Achieving a fast and quantitative EI-XCT will increase the potential for EI-XCT to be industrially deployed.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan
• Fellow: Huyge Ben

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

CLIMPACTH scopes safeguarding our built heritage in the context of climate change by combining state-of-the-art expertise on material characterization, degradation processes, hygrothermal modelling, climate modelling and built heritage. Our built heritage is an invaluable collection, being of exceptional cultural, social and economic importance. At the same time, it is also fragile and vulnerable due to its longterm exposure to the environment. Therefore, the sustainable management of our built heritage requires the implementation of actions to reduce its vulnerability and to increase its resilience to the risks of climate change. Furthermore, as the public and private sectors are pushed to invest in near-zero energy buildings for economic, ecologic and climate action reasons, the pressure on our built heritage to follow the same path is increasing, posing an additional threat to its historical and cultural values. Caring about our built heritage also means caring about the collections kept within, as historic houses, museums and churches are often located within built heritage. At the international level, the EU-funded projects NOAH's ARK (2004-2007) and Climate for Culture (2009-2014) set the first monumental steps to study the direct and indirect effects of climate change on built heritage at the European level. In addition, the International Council on Monuments and Sites (ICOMOS) has recently established a working group on climate change and heritage to comprehensively scope the intersection between cultural heritage and climate change, identifying both the strengths and challenges in this regard. CLIMPACTH will further build on these research initiatives at the international level and maximize collaboration with research initiatives at the national level to address the needs to adopt a global management strategy following the preventive conservation approach. The processes underlying these risk factors can be understood as material degradation in response to their constant exposure to various climate factors. In addition to these research initiatives, it is imperative to advance our knowledge of the characterization of historic materials and projected climatic parameters to address the risk assessment in an overall management strategy. The former will be assessed through in-depth research on material characteristics and degradation processes, and on-site assessment and monitoring of building envelope conditions. This will allow us to comprehensively generalize the built heritage envelopes and their condition for hygrothermal modelling. The latter can be assessed by estimating the evolution of climate factors under climate change conditions and their uncertainty using an ensemble of climate projections at global, regional and urban scales. This will allow us to propagate the uncertainties in climate projection to the scale of heritage buildings using dedicated bias correction methods. It will also allow us to carry out an integrated assessment of the degradation of risks under climate change conditions. By examining the particular conditions of built heritage and its exposure to the outdoor climate, risk factors will be reassessed and assessment strategies will be developed, taking into account the impact of the exterior on the interior and prioritizing the main threats.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Within the field of security, Customs and Border inspection have not had breakthrough technological developments in the last 20 years, since the introduction of X-ray screening. The limits of these current technologies are accentuated by the increasing diversity and novelty in trafficking materials, tools and methods. These limitations combined with the growing needs of inspection and control call for a disruptive innovative solution. Wanting to move a step up from the existing planar scanning methods with limited material identification results, several studies have identified potential solutions focused on: - High energy 3D X-ray tomography - Neutron interrogation/photofission - Nuclear resonance fluorescence (NRR) While these show good results and performances, they also have several important drawbacks, which limits their possible uses. Moreover, these solutions do not have common technological bricks meaning they can only lead to separate disposals. The proposed MULTISCAN3D investigates a new all-in-one system whose purpose is to become simultaneously a user-friendly, flexible, relocatable solution offering high-quality information for: - Fast high energy 3D X-rays tomography (as first line) - Neutron interrogation/photofission (as second line) - Narrow gamma ray beam based NRR (as second line) MULTISCAN3D will start by investigating and defining needs and requirements, in a technologically-neutral way, with Europe's most prominent Customs Authorities which will be translated to technical specifications. The main body of the research will be focused on three parts, following which, lab validations and real-environment demonstration will be carried out. These three work areas are: - Laser-plasma based accelerators as X-ray sources - 3D reconstruction for multi-view configurations and data processing - Detectors and source monitoring At the same time complementary techniques with chemical and SNM identification capabilities will be investigated.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

A widely used non-destructive method for material inspection is computer vision using RGB cameras along with matching image analysis. This allows to spatially resolve inhomogeneities in the materials. Computer vision however is limited to the visual part of the electromagnetic spectrum (400-700 nm), while many of the chemical processes and mineral formations in materials have particular reflectance properties in the near infrared (NIR, 700-1000 nm) and the shortwave infrared (SWIR, 1000-2500 nm). Spectrometers, on the other hand, are able to resolve these properties in the spectral direction but can only provide point-based measurements. The main scientific objective of this project is to exploit and enhance the potential of hyperspectral imaging in the NIR and SWIR range for the characterization of heterogeneously mixed and compound materials. For this, we will develop hyperspectral image analysis methods with increasing levels of complexity, ranging from spectral indices, characteristic for minerals and compounds in the materials, over methods based on spectral libraries, to supervised spectral unmixing methods. The developed methods will be validated on data acquired from homemade mixtures (mineral mixtures, sands, powdered clays and mortar) and implemented on two real use cases, i.e. mineral detection and corrosion detection.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Terahertz (THz) tomography is an up and coming technology that uses electromagnetic radiation with terahertz frequency for tomographic imaging. Like X-rays, THz waves provide information about the interior of an object through interaction with the object. THz waves interact with many materials in different ways. They are absorbed in polar materials such as water, penetrate most packing materials (plastic, paper, ceramics, …) and are completely reflected by metal. In contrast to X-rays, there are no known negative effects of THz waves, making their application attractive for biomedical purposes as well as industrial inspection, non-destructive testing, material science and agro-food applications. The Gaussian THz beam however, diverges much faster than an X-ray beam and reflection and refraction effects play a dominant role, preventing the use of conventional X-ray reconstruction techniques. In this project, we focus on the development of prior-knowledge based iterative reconstruction techniques for THz tomographic data that model the physics of the THz image formation in the image reconstruction process, as opposed to performing pre- or post-processing steps. Such algorithms are nearly unexplored for THz imaging and can greatly increase the applicability of the technique through a substantial improvement in image quality.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan
• Fellow: Lumbeeck Lars-Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Q-INSPEX aims at the development of novel imaging and image processing protocols to non-invasively and quantitatively inspect objects and subjects. Core imaging technologies herein are X-ray, (near)-infrared, and TeraHertz imaging. These technologies are largely complementary to each other and can be used in different set-ups as (i) an R&D tool to measure specific characteristics of materials (e.g. food structures or polymers), (ii) as a quality control procedure implemented within an industrial setting (i.e. compatible with processing speeds) or (iii) in-field inspections of crops and infrastructure (e.g. corrosion). Furthermore, they can be applied in a wide variety of domains: additive manufacturing, composites, art objects, textiles, archaeology, crops, food, etc.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan
• Co-promoter: Janssens Koen
• Co-promoter: Scheunders Paul
• Co-promoter: Steenackers Gunther
• Co-promoter: Van der Snickt Geert
• Fellow: De Samber Björn
• Fellow: Levrau Elisabeth

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Many properties of food, plants or seeds that are relevant to process engineering or quality are related to microstructure. Insight in food microstructure is therefore essential to control the quality of food. In the food factory of the future, flexible and efficient processes require dedicated sensor technology and automated analysis methods. In this context, X-ray computed tomography (XCT) is gaining traction as a non-destructive method to produce extremely detailed images of both internal and external features. Current XCT based analysis of food has a number of limitations however: i) Many microstructural features of food remain invisible due to poor image contrast in soft matter. ii) Visibility and quantification of structure from absorption XCT images strongly depends on image resolution, while relevant sub-resolution size features often remain undetectable. iii) Quality control requires reliable detection and classification methods that should be compatible with process line speeds and dedicated instrumentation that is currently out of reach to the food industry. With phase contrast XCT, images can be acquired with unprecedented contrast far surpassing conventional XCT contrast. This technique was only available at large-scale synchrotron facilities, but recent developments now allow for low brilliance, polychromatic X‐ray sources in lab XCT systems. The applicability to food analysis is however to a large extent unexplored and the 3D inline application is hindered by the long acquisition time. The aim of this project is to overcome these limitations by developing novel (inline) XCT phase contrast acquisition, reconstruction and inspection algorithms specific for the food industry. This will enable us to address issues such as limited visibility of microstructural features, non-detection of sub-resolution size features and incompatibility of reliable detection and classification methods with process line speeds.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Neuropathological research is an interdisciplinary field, in which imaging and image-guided interventions have become indispensable. However, the rapid proliferation of ever-more inquisitive technologies and the different scales at which they operate have created a bottleneck at the level of integration, a) of the diverse image data sets, and b) of multimodal image information with omics-based and clinical repositories. To meet a growing demand for holistic interpretation of multi-scale (molecule, cell, organ(oid), organism) and multi-layered (imaging, omics, chemo-physical) information on (dys)function of the central and peripheral nervous system, we have conceived μNEURO, a consortium comprising eight established teams with complementary expertise in neurology, biomedical and microscopic imaging, electrophysiology, functional genomics and advanced data analysis. The goal of μNEURO is to expedite neuropathological research and identify pathogenic mechanisms in neurodevelopmental and -degenerative disorders (e.g., Alzheimer's Disease, epilepsy, Charcot-Marie-Tooth disease) on a cell-to-organism wide scale. Processing large spatiotemporally resolved image data sets and cross-correlating multimodal images with targeted perturbations takes center stage. Furthermore, inclusion of (pre)clinical teams will accelerate translation to a clinical setting and allow scrutinizing clinical cases with animal and cellular models. As knowledge-hub for neuro-oriented image-omics, μNEURO will foster advances for the University and community including i) novel insights in molecular pathways of nervous system disorders; ii) novel tools and models that facilitate comprehensive experimentation and integrative analysis; iii) improved translational pipeline for discovery and validation of novel biomarkers and therapeutic compounds; iv) improved visibility, collaboration and international weight fueling competitive advantage for large multi-partner research projects.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Baets Jonathan
• Co-promoter: Cras Patrick
• Co-promoter: De Vos Winnok
• Co-promoter: Giugliano Michele
• Co-promoter: Kumar-Singh Samir
• Co-promoter: Staelens Steven
• Co-promoter: Stroobants Sigrid
• Co-promoter: Timmerman Vincent
• Co-promoter: Timmermans Jean-Pierre
• Co-promoter: Verhoye Marleen
• Co-promoter: Weckhuysen Sarah

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Visualizing nanoparticles at 3D atomic resolution using scanning transmission electron microscopy (STEM) is challenging. Current methods have specific shortcomings: STEM ADF tomography makes it hard to resolve light atoms and distinguish different atom types, and EDX tomography adds element specificity but is too slow and of too low resolution. We aim to resolve this issue by combining multi-tilt, through-focus, STEM imaging using multiple sensor signals: ADF and iDPC. For this purpose, new algorithms need to be developed that combine all information into a single 3D reconstruction, using a combination of physical models and, e.g., machine learning techniques. In this work package, we take advantage of the availability of both iDPC and ADF data as developed in the MSCA ITN Mummering project. First, model-based approaches will be considered to improve the atomic reconstruction by simultaneously reconstructing both datasets as a single atomic reconstruction. The aim of the model-based approach is primarily to exploit the availability of both datasets. Second, deep learning will be considered to take advantage of both datasets by integrating such network in the reconstruction in order to distinguish between different elements.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Hosseinnejad Shiva

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Electrical impedance imaging using high-density micro electrode arrays (HD-MEAs) is an emerging non-invasive technology to monitor cell cultures. The intention of this project is to develop a practical electrical impedance tomography (EIT) strategy for 3D imaging of cells cultured on 2D HD-MEAs.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Atom probe tomography is an analysis tool in materials science that allows to inspect the 3D chemical composition of needle shaped samples at the nano scale. The method works by field-induced evaporation. Ions are then consecutively emitted from the apex of the needle and are absorbed by a position sensitive detector. The result is a tomographic, atomically resolved image of the evaporated volume, represented as a point cloud in which each point is an atom. The current reconstruction approaches however were developed with homogeneous samples in mind and do not account for the complex shape of the sample surface, which evolves during the field evaporation process. The goal of this project is to develop new reconstruction methods that take the shape into account.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Analysis of patients scans and healthy control pre and -post 3D scans with quantitative measures of difference. Optical scans are acquired from African subjects with a specific skin disease and there progression is quantified through image analysis of a series of scans.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Magnetic resonance imaging (MRI) is a medical imaging technique that generates excellent soft-tissue contrast and allows for investigating both anatomy and function of tissues noninvasively. In conventional MRI, direct HR acquisition requires long scan times to achieve adequate precision and spatial resolution of the resulting MR image. From a diagnostic perspective, long scan times increase the likelihood of motion artefacts, whereas, from an economical perspective, they reduce the throughput. In addition, long scan times cause discomfort for patients. Multi-slice super-resolution reconstruction (MS-SRR) has the potential to reduce this limitation, improving the inherent trade-off between resolution, SNR, and scan time. MS-SRR consists in estimating a 3D high-resolution (HR) image from a series of 2D multi-slice images with a low through-plane resolution. Two strategies are conventionally adopted to acquire data for an MS-SRR experiment. The first consists in acquiring a set of multi-slice images with parallel orientations, where each image is shifted in the through-plane direction by a different, sub-pixel distance. The second consists in acquiring rotated multi-slice images, where each image is rotated around the frequency and/or phase encoding axis by a different rotation angle. These two strategies will be compared in terms of accuracy and precision of the reconstructed images. MS-SRR estimation is generally an ill-posed problem and the use of regularization has an impact on the SRR estimated image. I will investigate a Bayesian SRR framework in which local correlation information is learnt from MRI images and used to stabilize the SRR estimate. An optimal experimental design framework will be developed in which the Bayesian Mean Squared Error (BMSE) of the MAP estimator is proposed as a performance criterion, to compare the two aforementioned acquisition strategies in the context of regularized MS-SRR. We plan to validate the BMSE-based predictions on simulated and real data. Finally, we aim to extend the MS-SRR optimal experimental design framework to quantitative SRR (qSRR). In qSRR, a high resolution (HR) relaxation parameter map is estimated from a series of weighted multi-slice images with a low through-plane resolution. Each slice of each LR image can be acquired with different weighting settings, thus offering maximum flexibility to optimize the weighting settings for each slice individually. Aiming at the highest attainable precision for a given acquisition time, we will optimize the experiment design of the SRR framework by searching for the optimal acquisition parameters. This research is expected to further improve the trade-off between signal-to-noise, resolution, and scan time in qSRR, by for example allowing precise estimation of HR parameter maps from shorter scans.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Nicastro Michele

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

Diffusion magnetic resonance imaging is a powerful, non-invasive technique to investigate the microscopic properties of tissues, based on analyzing the diffusion of water molecules, which is influenced by the tissues' microstructure. However, the clinical application of high resolution diffusion imaging is impeded due to its long scanning time. To reduce scan time, fast acquisition schemes such as multi-shot acquisition and undersampled data acquisition have been introduced. However, these acquisition schemes may introduce serious artifacts in the reconstructed diffusion parameter maps if not complemented with smart image reconstruction. In this project, we introduce an advanced reconstruction framework that allows accelerated imaging by varying the diffusion contrast settings during the acquisition of a single image, e.g. for each shot in the multi-shot acquisition, introducing intra-scan modulation. This model-based reconstruction framework estimates diffusion parameter maps directly from the acquired intra-scan modulated data and simultaneously corrects for artifacts related to shot-to-shot phase inconsistencies. By now, the statistical performance of this framework has been assessed in Monte Carlo simulation studies. In the next phase of the project, the framework will be extended to include higher-order phase patterns to account for more complex subject motion. In addition, the framework will be combined with common acceleration methods such as parallel imaging to aim for higher acceleration rates. Finally, we aim to define the optimal imaging settings including sampling strategy and optimal diffusion contrast set-ups using a statistical experiment design based on a Cramér-Rao Lower Bound analysis. The feasibility of this approach will be investigated in real-data experiments considering at first, retrospective acceleration and, secondly, direct acquisition of multi-shot intra-scan modulated data.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Shafieizargar Banafshe

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

In mineralogical studies, the use of hyperspectral images helps to detect different minerals in a faster and more accurate manner. However, it can be problematic to process such data, as the acquisition of sufficient ground truth information is challenging. Therefore, the main goal of this research is to develop unsupervised learning techniques (sparse subspace-based clustering algorithms) to distinguish different mineralogical features from hyperspectral images. As a first milestone for the proposed research topic, a new sparse subspace-based clustering algorithm was developed to cluster the mineralogical features into meaningful groups. The proposed algorithm is able to process highly mixed and complex data in a robust and fast manner. As a follow-up, spatial information was exploited in a newly developed clustering algorithm. Such information helps to take into account spatial structures of mineralogical samples. By including the spatial information, the precision of the proposed algorithm increased. In the period to come, we will explore the potential of deep clustering methods, based on state-of-the-art sparse autoencoder-based algorithms, or the concepts of deep image priors. As a result of the study, fast and robust deep learning-based clustering algorithms will be developed, to analyze complex hyperspectral images from mineralogical and geological features.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Rafiezadeh Shahi Kasra

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

As defined by the Quantitative Imaging Biomarkers Alliance (QIBA), quantitative imaging aims at extracting "quantifiable features from medical images for the assessment of normal or the severity, degree of change, or status of a disease, injury, or chronic condition relative to normal". However, the lack of widespread consensus and integration in commercial software of quantitative MRI (qMRI) methods have hampered both the direct comparison between results of different research groups as well as the translation of cutting-edge qMRI technology to the clinic. The general aim of this research project is to bridge the gap between qMRI research and clinical applications using the syngo.via Frontier platform from Siemens Healthineers. This platform serves as an integrated research environment for advanced post-processing of medical images, allowing for both the development and the evaluation of algorithms in close collaboration with clinicians. An established group of MRI post-processing algorithms, commonly referred to as super-resolution reconstruction (SRR) techniques, are used to estimate a high-resolution image from an acquired set of low-resolution images, thereby improving the MRI trade-off between signal-to-noise ratio (SNR), spatial resolution and scan time. Specific SRR methods have been developed for high-resolution anatomical MRI, but also for qMRI by integrating quantitative models that enable the estimation of biophysical parameters for tissue characterisation. Although SRR holds applications in a variety of clinical fields, its clinical potential in the context of musculoskeletal (MSK) MRI remains to be thoroughly investigated. Consequently, the specific aim of this research project is twofold: 1. Following the demonstrated feasibility of SRR TSE MRI of the knee, we aim to evaluate the clinical application of the described anatomical SRR technique for accelerated high-resolution isotropic 3D knee MRI by comparison with the current clinical standard. Furthermore, the integration of the SRR post-processing algorithm for MSK MRI on the Siemens syngo.via Frontier platform will be finalized to facilitate clinical evaluation. 2. As previously reported, 3D UTE Spiral VIBE MRI shows great promise for fast T2* mapping of knee structures. To further improve accuracy and precision of the T2* estimation, we aim to develop a quantitative SRR framework for rapid isotropic T2* mapping of the knee, based on both ultra-short echo time (UTE) and multi-echo gradient echo (MEGE) imaging. In light of QIBA's mission, the developed quantitative SRR framework will be used to probe the suitability of the biophysical short and long T2* parameters as biomarkers of MSK tissue structural integrity. More specifically, the framework will be used to assess the severity of anterior cruciate ligaments (ACL) injuries and to evaluate the healing process of reconstructed/repaired ACLs.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Multiple Sclerosis (MS) is an autoimmune inflammatory disease of the central nervous system. There is no cure for MS, but many treatments have been developed to slow down its progression. Disease progression monitoring and clinical decision making often rely on the expanded disability status scale (EDSS). Unfortunately, EDSS suffers from poor reliability, repeatability, and high inter-rater variability. A first goal of this project is to reduce the inter-rater variability and increase repeatability in quantifying the risk of patient disability by developing a machine learning technique based on anatomical magnetic resonance images (MRI) and diffusion MRI (dMRI). As a first step, we will focus on the prediction of the EDSS scoring, but other clinical scores will be included as well. To develop an automated EDSS scoring method, a large database is required. Such databases are typically composed of images from multiple centers, and hence depend on scanner hardware, reconstruction algorithms and acquisition protocols. These factors lead to high intra- and intersite variability in structural MRI data, and even more in parameters derived from dMRI data. A second goal is to develop, implement and validate harmonization methods for structural and dMRI data, to reduce unwanted variability while preserving biological variability. Working towards this goal, I co-authored a review paper on dMRI harmonization methods [Pinto, et al. 2020]. A next step is to validate a recently proposed diffusion harmonisation method "Method of Moments" [Huynh, et al. 2019] on in vivo dMRI data. Finally, as part of the Horizon 2020 initial training network B-Q MINDED, the project's ultimate goal is the integration of the harmonisation and EDSS scoring algorithms in a product that can be used in clinical trials and, in a later stage, in the daily clinic. Roadmap September 2021-October 2021: Finalizing of the EDSS scoring application based on anatomical MRI. Submission of a journal manuscript "Prediction of EDSS scores in MS patients from MRI" by the end of October 2021. November 2021-December 2021: Finalizing implementation and validation of deep-learning approaches for the harmonisation of anatomical and diffusion MR images. January 2022 - February 2022: Automated EDSS scoring based on harmonized structural and dMRI data. March 2022-July 2022: preparation of the PhD thesis. References Pinto, M.S., Paolella, R.,…..et al. "Harmonization of brain diffusion MRI: Concepts and methods." Frontiers in Neuroscience 14 (2020). Huynh, Khoi Minh, et al. "Multi-site harmonization of diffusion MRI data via method of moments." IEEE transactions on medical imaging 38.7 (2019): 1599-1609.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Diffusion-weighted magnetic resonance imaging is a non-invasive technique to reveal the brain's microstructural properties by probing the local diffusion of water molecules. Fitting mathematical models to diffusion MRI data allows to extract quantitative information and, among these models, the diffusion tensor imaging (DTI) model is the most commonly applied. However, recent literature has shown that the diffusion kurtosis imaging (DKI) model can provide more accurate estimates of diffusion tensor properties as well as additional information in clinical applications. Unfortunately, the quality of the diffusion metrics extracted from these models is degraded by several acquisition artefacts, such as Gibbs ringing, eddy current distortions and susceptibility-induced artefacts. Besides these well-known artefacts, voxels in DW images may suffer from additional problems: ● signal intensity outliers resulting from motion, cardiac pulsation or system instabilities can compromise the parameter estimates to an extent that they are no longer useful; ● image voxels are relatively large (2 to 3 mm isotropic) and thus susceptible to partial volume effects, which is particularly a problem in brain images when cerebrospinal fluid contamination occurs making the interpretation of diffusion markers ambiguous and no longer tissue-specific. A first aim of this project is to improve and validate an outlier-robust framework for diffusion and kurtosis parameter estimation. During the initial phase of the PhD project, the performance of such a framework was assessed in simulation experiments, thereby ignoring spatial correlations of outliers. As a logical step for improving the method, prior information on how outliers correlate within a slice will be included. Subsequently, a validation study will be performed to assess the reproducibility of DKI metrics in real test-retest datasets. In the second and third year of the PhD project, an advanced bi-compartment model based on the combination of diffusion and relaxometry data has been proposed for correcting free water contamination in multi-shell multi-echo diffusion data. This work has resulted in a journal paper that will be submitted in the second quarter of 2021. This promising approach exploits the combination of diffusion and relaxometry data as a rich source of information, but is not applicable to datasets acquired with a single echo time, which are typically acquired in clinical practice. For this reason, our next research goal is to implement and validate approaches for partial volume correction in single/multi-shell single-echo acquisitions. For this purpose, the potential of artificial intelligence solutions will be explored to deal with the ill-conditioned parameter estimation problem. Finally, as part of the Horizon 2020 initial training network (ITN) B-Q MINDED, the ultimate aim of the project will be to integrate the developed techniques in a regulatory approved quantitative MR product that can be used in clinical trials and, in a later stage, in daily clinical practice for improved assessment of drug efficacy and patient follow-up.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main motivation of this project is a major opportunity that came up in the beginning of 2020, when 3 hyperspectral cameras and a spectrometer became available to our lab. This opportunity will enable us to generate high quality ground reference data for the validation of the nonlinear unmixing methods that are developed in the GEOMIX project. The objective of this support project is the acquisition, calibration and analysis of ground truth datasets for nonlinear spectral unmixing, through very well thought-out experiments in a controlled lab environment.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In X-ray computed tomography (XCT), X-ray images of a sample are taken from multiple angles and used to form a 3D reconstruction of the full sample, including many internal features. In a recently rising field in XCT, called phase contrast CT, a specialized set-up is used to obtain a signal that not only holds information on the absorption of the X-rays (as in traditional XCT), but also on the local scattering power in the sample and on the phase shift, a wave property. In the standard phase contrast reconstruction workflow, the acquired data is first separated in an attenuation, differential phase and dark field signal. These signals are then separately reconstructed, using an algorithm derived from traditional XCT, after which the data of the different signals is evaluated as a whole. We focus on two problems in this workflow. First, the signal separation and reconstruction use a linear model, which often does not align with reality. This model assumes a source that sends a single type of X-ray, whereas in a general setting there is a whole spectrum. Secondly, there is a relation between the different signals that are reconstructed, as they all come from the same sample. Currently this is not exploited during the reconstruction. The end goal of this project is to create a model for reconstruction exploiting all phase contrast modalities at the same time, while accounting for the different X-ray energies, such that phase contrast can be used in a quantitative setting.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan
• Fellow: Six Nathanaël

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Using X-ray Computed Tomography (XCT), internal and external characteristics of an object can be visualized in 3D in a non-destructive manner. Medical applications of XCT are well known, but also in other industrial sectors the possible applications of XCT are numerous, such as material characterisation, process and quality control and safety inspection. However, conventional XCT does not penetrate the industry very well, partly because each industrial application requires a specific XCT scan, processing speed and quality, and the XCT equipment available on the market is hardly adapted to these needs. After all, the majority of available CT equipment is very rigid in terms of scan geometry and not cost-efficient. In order to put new, innovative X-ray scanning methods into practice, imec-Visionlab purchased a custom-made X-ray device in 2019: the UniTomXL (Tescan-XRE). The alternative name provided for this device, FleXCT, emphasizes its unprecedented flexibility in terms of possible X-ray geometries for Industrial applications. Moreover, over the past 10 years, imec-Visionlab has developed the ASTRA toolbox with which the recorded X-ray scan can be reconstructed into detailed 3D images. However, in order to quickly respond to industrial XCT requests via an efficient service platform, a high-performance workflow is needed, consisting of 1) FleXCT initialization, 2) FleXCT scanning, 3) 3D image reconstruction, 4) Image visualization and analysis. In this project, the focus will therefore be on the development of such a workflow from customer demand to analysis. For this purpose, we will develop new scanning scripts, seamlessly link the ASTRA toolbox reconstruction algorithms to these scripts, and realize fast visualization and analysis of the reconstructed 3D models via the DragonFly software package (ORS, Canada, www.theobjects.com/dragonfly). Thanks to this new workflow, an efficient service platform will be offered to both academic and industrial partners.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

The project aims to improve print quality and reduce waste and cost by in-line real-time monitoring of the melt pool and the product during printing and controlling printing in-the-loop. It aims to produce the first off-axis system based on video analysis.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The major scope of this project is the evaluation of the diagnostic benefit of DFI. Therefore, the project is structured in two parts. The first track is the development of a DFI prototype for large field of view including software algorithms to reconstruct the absorption and dark field images. A second track is the clinical assessment to finally test DFI with specimen (ex- and in-vivo) to evaluate the diagnostic value of this new technique. An essential aspect of the project is the design and development of a prototype system that will be usable in clinical routine and which allows production at moderate cost. This requires fundamental changes of state-of-the-art PCI designs (hardware & software). The test-prototype which will be used for a pre-clinical trial will include these design changes to get significant evidence of the diagnostic value of the new DFI system.

Researcher(s)

• Promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Fiber reinforced polymers (F s) are increasingly used in critical components in the aerospace and automotive industry because of their low weight, strength, and cost effectiveness. Construction of F s requires an in-depth understanding of their microstructure to evaluate the strength and integrity of the composites. High resolution X-ray computed tomography has become the method of choice to investigate the composition and internal structure of F s. Unfortunately, conventional attenuation based X-ray imaging suffers from poor spatial resolution and contrast between the fibers and the polymer matrix. Fortunately, imaging methods have recently become available for lab-X-ray systems that allow to measure the local X-ray scattering (dark field imaging), leading to images with unprecedented contrast complementary to the conventional attenuation contrast. Dark field X-ray imaging is especially useful to image F s as it allows to reconstruct the full scattering profile in each voxel. However, crossing or intertwined fibers within a voxel are hard to disentangle, which makes quantification of distributions of fiber directions challenging. In this project, we will develop new models for superresolution dark field X-ray imaging that allows to quantify F fiber distributions with a subvoxel spatial resolution. This may lead to a better understanding of F properties and ultimately a better design of such materials.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In X-ray computed tomography (XCT), X-ray abso tion images of a sample are taken from multiple angles and subsequently used to form a 3D reconstruction of the full sample, based on the attenuation of X-rays. In a recently rising field in XCT, called edge illumination phase contrast CT, a specialized set-up is used to measure, apart from the attenuation, also the local scattering power in the scanned sample and the phase shift of the X-rays. Compared to attenuation, the scatter and phase signals hold complementary information of the scanned sample. Since these signals cannot be measured directly, an absorbing mask (a grating) must be placed in front of the sample and another mask in front of the x-ray detector. In the standard phase contrast imaging workflow, these masks are custom made for a specific imaging geometry and perfectly aligned to each other to achieve the right measurement conditions. The main drawback of this rigid set-up is that geometry changes that are common practice in traditional CT (e.g. zooming in on a sample to optimize the resolution and field-of-view) are not possible. Our aim here is to overcome this limitation by designing novel masks that adapt to geometry changes of the XCT set-up. This fundamental change will open up phase contrast imaging to a much larger variety of sample sizes and at different scales of resolution.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

A VLAIO COOCK project on novel applications within X-ray CT to inspect different types of materials and objects. MicroCT is a powerful, non-destructive technique for producing high quality 3D images of objects based on a set of X-ray projections. The main aim of the project is to define specific use cases that can be explored using our X-ray CT system (FLEX-CT) within an industrial setting.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Foams are found worldwide in a huge array of products, ranging from food to polyurethane foam (PU foam). However, the physics underpinning the foam formation process is not yet fully understood. A versatile and popular technique to investigate foam structure is micro X-ray computed tomography (microCT). MicroCT is a powerful, non-destructive technique for producing high quality 3D images of static objects based on a set of X-ray projections. In order to visualize the dynamics, a series of subsequent 3D images is traditionally acquired. This approach assumes the object to remain still during the acquisition of a single 3D image. However, in most dynamic imaging situations, this is only approximately valid. Therefore imaging of a fast dynamic processes such as foam formation is currently limited to synchrotron light sources as they are able to acquire a 3D image in the order of a few seconds. Unfortunately, synchrotron beamtime is very limited and experiments are typically queued for 3 to 12 months. This project will therefore focus on improving the image quality of lab-based microCT experiments of PU foam by developing, multimodal (absorption and phase data) 3D and 4D reconstruction algorithms. The key novelty lies in the use of specific prior knowledge about the foam cell shape and its material properties. On the application side my research will facilitate lab experiments and thereby greatly reduce the experiment cycle time in the industry.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan
• Fellow: Renders Jens

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Terahertz (THz) tomography is an up and coming technology that uses electromagnetic radiation with terahertz frequency for tomographic imaging. Like X-rays, THz waves can give information about the interior of an object through interaction with the object. In contrast to X-rays, there are no known negative effects of THz waves, making their application attractive for biomedical purposes as well as industrial inspection. THz rays interact with many materials in different ways. They are absorbed in polar materials such as water, penetrate most packing materials (plastic, paper, ceramics, …) and are completely reflected by metal. With terahertz tomography it is possible to, for instance visualize the contents of a sealed package. The Gaussian THz beam however, diverges much faster than an X-ray beam and reflection and refraction effects play a dominant role, preventing the use of X-ray reconstruction techniques. Here, we focus on the development of prior-knowledge based iterative reconstruction techniques for pulsed terahertz tomographic data that not only provides information on the absorption but also on phase differences.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Fellow: Lumbeeck Lars-Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

There is a widespread rising need from industry to move towards 100% non-destructive inline inspection and quality control. The main challenge in X-ray based inspection is to go beyond classical X-ray radiography image processing and make the step towards fast and robust 3D inspection. This challenge is rooted in the difficulty of disentangling the 3D spatial information that is encrypted in the X-ray radiographs. imec-Vision Lab has developed methodology that can enable the introduction of high throughput inline tomography for industrial quality control. In this project we aim to push this technology past TRL4 through the development of a computationally efficient and more robust software platform, which can greatly facilitate the creation of a spin-off.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Advanced composite materials (ACMs) typically contain two or more constituents, such as matrix, fibers, and pores, with different physical and chemical characteristics. When combined, they produce a material with unique properties in terms of weight, strength, stiffness, or corrosion resistance. To inspect and study their 3D internal structure in a non-destructive way, the ACMs are imaged with X-rays, after which a 3D image is reconstructed from the X-ray radiographs, and further processed and analyzed in multiple sequential steps. This conventional workflow, however, suffers from inaccurate modeling and error propagation which severely limits the accuracy with which ACM parameters of interest can be estimated. In this project, we will develop a paradigm shifting approach in which the quantification of ACM parameters is substantially improved. This will be realized in a novel workflow by 1) accurately modelling all constituents of the ACM (matrix, pores, and fibers); 2) directly estimating the ACM model parameters from the X-ray radiographs, thereby preventing error propagation by providing a feedback mechanism; 3) analyzing the workflow's parameter space with respect to sensitivity and stability of parameters of interest. In this project, we develop methods that quantify ACM parameters, by targeting a new workflow for 1) accurately modeling all components of the ACM (matrix, pores and fibers); 2) estimating directly the parameters of the ACM model of the X-rays, thus preventing error propagation.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Elberfeld Tim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Clothing webshops have to deal with a large number of returns because the customer orders the wrong size, so a lot of money (€600 billion worldwide) is lost. SHASIZE aims to create a true-to-life virtual mannequin, based on a few simple input parameters (length, weight, circumferences). This mannequin determines how well the garment fits.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Fibrous materials are found in biology (e.g. skin, muscle, tendon, ...), but also in industry in the form of composite materials in critical components of the aerospace, automotive and building applications. Not surprisingly, there is a great demand, both clinical and industrial, for an in-depth understanding of the microstructural response of these fibrous materials to external loading parameters defining their elasticity, strength and structural integrity. In this project, a novel experimental 4D characterization toolbox based on X-ray computed tomography (XCT) will be developed, including non-invasive contrast agents and dedicated in situ measurement devices, along with advanced 4D image reconstruction and analysis methods and computational models. Two representative case studies will demonstrate the general applicability of our approach: 3D printed fibre reinforced composites and biological tissues. The proposed 4D characterization approach will allow us to gain crucial insight into the microstructural changes that occur during dynamic functional testing of both types of fibrous materials. In turn, the improved knowledge of the dynamic material behaviour can pave the way towards optimized design and production of novel 3D printed composite materials and towards a more intelligent design of next-generation solutions for tissue restoration and regeneration. The project brings together a multidisciplinary team of experts from three Belgian universities, and will facilitate the translation of the developed 4D characterization toolbox, as well as the individual methodologies, towards industry, hospitals and research centers.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Atom probe tomography is an analysis tool in materials science that allows to inspect the 3D chemical composition of needle shaped samples at the nano scale. The method works by field-induced evaporation. Ions are then consecutively emitted from the apex of the needle and are absorbed by a position sensitive detector. The result is a tomographic, atomically resolved image of the evaporated volume, represented as a point cloud in which each point is an atom. The current reconstruction approaches however were developed with homogeneous samples in mind and do not account for the complex shape of the sample surface, which evolves during the field evaporation process. The goal of this project is to develop new reconstruction methods that take the shape into account.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Terahertz radiation is non-ionizing and can be used for 3D inspection. In this project, new reconstruction methods are developed for Terahertz tomography. The THz beam is modelled and incorporated into iterative reconstruction methods.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Conventional analytical methods for CT reconstruction result in artifacts for non-ideal, constrained, CT acquisitions such as when only a limited angular range is available, or when only few projection images can be acquired due to time constraints and when the image formation model is simply inadequate. I will focus on the development of novel reconstruction techniques that take advantage of prior knowledge (e.g. sample shape, materials, energy spectra) in both X-ray absorption and quantitative phase contrast tomography to solve these issues and I will interconnect the developed algorithms with equally flexible image acquisition hardware, thereby taking advantage of prior knowledge of the object to be scanned and as well as of the imaging hardware itself.

Researcher(s)

• Promoter: De Beenhouwer Jan
• Fellow: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Multi-tissue spherical deconvolution of diffusion MRI (dMRI) is a popular analysis method that provides the full white matter fiber orientation density function as well as the densities of cerebrospinal fluid and grey matter tissue in the living human brain, completely noninvasively. It can be used to track the long-range connections of the brain and provides a tract-specific biomarker for neuronal loss in the study of neurodegenerative diseases. Currently, the technique can be regarded as a macroscopic approach: it breaks up the dMRI voxels in terms of tissues rather than cellular components, the latter being potentially more relevant biomarkers. Unfortunately, recent studies have demonstrated that conventional low-dimensional dMRI scans lack the information to resolve these microstructural features. In this proposal, I will take multi-tissue spherical deconvolution to the next (microscopic) level by leveraging high-dimensional dMRI scans. These next-generation scans have shown great promise to disentangle different microstructural compartments. The new multi-compartment spherical deconvolution approach will allow simultaneous estimation of a high quality axonal orientation density function as well as the densities of cell bodies and extracellular space. This will enable high-quality fiber tracking and at the same time provide more relevant biomarkers, and will help spherical deconvolution to maintain its position as one of the go-to tools for dMRI analysis.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Jeurissen Ben

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In X-ray computed tomography (XCT), X-ray images of a sample are taken from multiple angles and used to form a 3D reconstruction of the full sample, including many internal features. In a recently rising field in XCT, called phase contrast CT, a specialized set-up is used to obtain a signal that not only holds information on the absorption of the X-rays (as in traditional XCT), but also on the local scattering power in the sample and on the phase shift, a wave property. In the standard phase contrast reconstruction workflow, the acquired data is first separated in an attenuation, differential phase and dark field signal. These signals are then separately reconstructed, using an algorithm derived from traditional XCT, after which the data of the different signals is evaluated as a whole. We focus on two problems in this workflow. First, the signal separation and reconstruction use a linear model, which often does not align with reality. This model assumes a source that sends a single type of X-ray, whereas in a general setting there is a whole spectrum. Secondly, there is a relation between the different signals that are reconstructed, as they all come from the same sample. Currently this is not exploited during the reconstruction. The end goal of this project is to create a model for reconstruction exploiting all phase contrast modalities at the same time, while accounting for the different X-ray energies, such that phase contrast can be used in a quantitative setting.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan
• Fellow: Six Nathanaël

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

PC-CT reveals complementary information to traditional attenuation based X-ray imaging (i.e. higher contrast in soft tissue). The FleXray system will allow us to acquire data to fully explore a far wider range of applications and opportunities for PC-CT that are currently not possible: ● Exploration of advanced CT acquisition models to enable reconstruction from (1) fewer projection images and (2) projection images acquired during continuous sample rotation. This will result in faster PC-CT imaging (currently up to 8 times longer than regular CT). ● Dark field tomography is only in its infancy but recently showed huge potential in material characterisation. The FleXray system will open new research lines on dark field tomography, in particular in accurate and precise estimation of localized scattering profiles. ● Development of Krylov solvers with much faster convergence for simultaneous multimodal reconstruction of full 3D images of attenuation, phase and dark field signals.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan
• Co-promoter: Janssens Koen
• Co-promoter: Vanroose Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The ASTRA toolbox is an open source platform for tomographic reconstruction. In this project, extensions for the ASTRA toolbox have been developed. These include refractive imaging such as TeraHertz tomography.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Magnetic resonance imaging (MRI) is one of the most useful and rapidly growing neuroimaging tools. Unfortunately, signal intensities in conventional MRI images are expressed in relative units that depend on scanner hardware and acquisition protocols. While this does not hinder visual inspection of anatomy, it hampers quantitative comparison of tissue properties within a scan, between successive scans, and between subjects. In contrast, advanced quantitative MRI (Q-MRI) methods like MR relaxometry or diffusion MRI do enable absolute quantification of biophysical tissue characteristics. Evidence is growing that Q-MRI techniques detect subtle microscopic damage, enabling more accurate and early diagnosis of neurodegenerative diseases. However, due to the long scan time required for Q-MRI, causing discomfort for patients and limiting the throughput, Q-MRI methods have not entered clinical practice yet. B-Q MINDED aims to overcome the current barriers by developing widely-applicable post-processing breakthroughs for accelerating Q-MRI. The originality of B-Q MINDED lies in its ambition to replace the conventional rigid multi-step processing pipeline with an integrated single-step parameter estimation framework. This approach will unlock a wealth of options for optimization of Q-MRI. To accomplish this goal, B-Q MINDED proposes a collaborative cross-disciplinary approach (from basic MR physics to clinical applications) with strong involvement of industry.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: den Dekker Arjan
• Co-promoter: Guns Pieter-Jan
• Co-promoter: Verhoye Marleen

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

Motion patterns are crucial markers for the health of a horse. The lack of accurate motion analysis systems leads to subjective diagnoses with a negative effect on the treatment outcome. The current most accurate systems are based on X-ray radiographs of the subject during motion. The 3D reconstruction from the planar images requires a prior CT-scan of the subject. The techniques are not suitable for medical applications because they require placement of invasive markers into the subject and require a lot of manual processing. Moreover, they can not handle deformable motions. As a result, the cushions on horse hoofs which deform during landing, can not be imaged. We propose a motion reconstruction technique that replaces the CT-scan by a statistical shape and intensity model. Omitting a CT-scan reduces examination costs, time and radiation dose. A statistical model describes the variation in shape and densities present within the population. To describe moving subjects, such a model needs to be articulated over time. This will be extended with a model for the deformable dynamics of soft tissues. The motion reconstruction technique will autonomously find the right shape and pose of the model based on the X-ray images by comparing them with a simulated X-ray image of the model. The novel technique will serve as an objective diagnostic tool for diagnosis, follow-up and validation of innovative orthopedic products by means of motion analysis, both for animals and humans.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Van Houtte Jeroen

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Atom probe tomography (APT) is a chemical analysis technique that provides a three-dimensional atom distribution of a measured specimen. A sharpened specimen is placed into a vacuum chamber and aligned to the center of an ion detector with a high voltage bias applied between the tip and the detector. A high electric field (about > 10 V/nm) is then formed at the apex of the tip, while the atoms at the surface of the apex are ionized and the intensity of the electric field is close to the threshold of breaking atomic bonds. For analyzing low conductive materials, a continuous pulsing laser is commonly introduced as a supplement of the thermal energy which helps ions at the apex to overcome the energy barrier of evaporation. Evaporated ions are detached from the tip surface and are accelerated toward the detector according to the electric field distribution between the tip and the detector. The impact position on the detector and the travelling time, as named time-of-flight (TOF), from emission to detection are measured. It is noteworthy that, with the limited size of a detector, only those ions in the field-of-view (FOV) will reach the detector. Moreover, because of the detector efficiency, only 50-70% of the ions that reach the detector will be recorded. These effects cause significant uncertainties on determining the volume for a reconstruction. In this project we will develop novel reconstruction methods for APT.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main objective of the project is the creation of a powerful yet flexible hyperspectral unmixing methodology, based upon sound physical and mathematica! principles, validated on extensive artificial and real data sets, and made publicly available through publication and sharing of program code. The unmixing methodology is based upon the recently developed multilinear mixing model, and will extend this state-of-the-art method with several additional capabilities.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Dynamic plantar pressure imaging (PPI) refers to the measuring, across time, of pressure fields between the foot and the ground. PPI is used, in part, to diagnose foot problems such as metatarsalgia and plantar fasciitis. Despite the widespread clinical use of PPI, its diagnostic potential has not been fully exploited. PPI creates large and dynamic datasets that cannot be easily analysed and interpreted by the human brain. As a result, PPI images are subsampled before being clinically examined, which discards potentially valuable information. The objective for this action is to improve the diagnostic value of PPI through the introduction of a computer-aided diagnosis (CAD) system called CAD WALK.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Booth Brian

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Advanced manufacturing techniques, often based on computer aided design (CAD) models, are transforming the industrial landscape and offer exciting opportunities for producing tailor-made products with high added-value. At the same time, specifications and quality standards of end products are stringent and therefore sophisticated inspection tools are needed. In an industry 4.0 perspective, inspection occurs preferably inline to enable a rapid remediation of disturbances causing material defects and/or dimensional deviations. Hence, there is a growing demand for fast and flexible 3D metrology solutions in the factories of the future. In this context, X-ray computed tomography (CT) is gaining traction as a non-destructive method to produce extremely detailed images of both internal and external features of complex objects. However, conventional CT inspection approaches typically require many (several hundreds) X-ray projection images from a large number of viewing angles and subsequently a full 3D image reconstruction is performed. This results in a number of limitations: i) due to the lengthy acquisition and reconstruction process, CT is typically performed for offline inspections and R&D activities. Real-time inline CT scanning to achieve a 100% dimensional metrology inspection rate is not possible with the current CT systems. ii) conventional CT systems have a rigid well-defined setup, i.e. requiring either that the object can be put inside the scanner or that the source-detector system can physically rotate 360° around the object. As a result, larger objects such as a wing of an airplane or a partly assembled car cannot be scanned. iii) 3D reconstructed images may suffer from numerous artefacts (due to misalignment, beam hardening, etc.) while the traceability and uncertainty of CT measurements for metrology applications is insufficiently documented. In this project, we propose a radical paradigm shift by breaking with the traditional X-ray 3D metrology workflow through developing a new framework for 3D metrology that addresses the above mentioned problems. If successful, this SBO project will result in a flexible X-ray metrology toolkit to enable fast inline QC during production and to perform inspection tasks of larger parts. The identification of hidden defects and deviations from the nominal geometry during production will help to produce high quality products, as efficiently as possible and with a minimum of waste.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Cardiac imaging plays an important role in the detection of pathologies of the heart, including coronary and valvular heart disease. It is also increasingly used for planning of complex surgery, and for the patient-specific fitting of medical implants such as artificial valves. Up till recent, dynamic imaging of the heart motion was limited to fast ultrasound (US) imaging, or MRI and CT limited to 2D or a reduced axial field of view. The advent of wide-area detectors with high tube rotation speeds has now enabled acquiring CT volumes covering the entire heart, several times per second. Dynamic or 4D (3D+T) CT is of great promise to clinical cardiac imaging. The modality is particularly suited for applications requiring image processing such as physics-based modelling, in which models of the anatomy are extracted from the image as a starting point for computer simulations. In comparison to US, CT offers a larger field of view and superior signal to noise ratio, making it far better suited for whole-heart segmentation and geometric modelling. Conversely, 4D US offers superior temporal resolution, and provides greater detail on fine structures such as heart valves. Combining dynamic CT with US, would allow benefiting from the advantages of both modalities, and could lead to a robust and accurate workflow for extracting detailed, patient-specific information on heart anatomy and motion. Inclusion of 4D models in physics-based modelling could bring such simulations to a new level of realism, enabling their use for planning of complex interventions and in-silico trials of cardiovascular devices affected by motion. In term, this will reduce the uncertainties associated with such interventions through more accurate device sizing and positioning, and accelerate the development of novel cardiovascular implants. The DIASTOLE consortium aims to develop a novel 4D workflow for performing physics-based simulations for cardiovascular procedures in a dynamic environment, using patient-specific parametric models of the heart and main arteries, obtained from dynamic CT and US.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Huysmans Toon

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Magnetic resonance imaging (MRI) is one of the most used neuroimaging techniques. Unfortunately, signal intensities in conventional MRI images are expressed in relative units that are dependent on hardware and software. This does not hinder visual inspection of anatomy, but severely complicates quantitative comparisons of the signal intensity within a scan, between successive scans, and between subjects. In contrast, MR relaxometry is an MRI technique that generates quantitative maps of absolute biophysical tissue characteristics (Deoni et al., 2010). Evidence is growing that MR relaxometry detects subtle microscopic tissue damage, which could lead to earlier diagnosis of various brain diseases including multiple sclerosis (Vrenken et al., 2006; Roosendaal et al., 2009 and Papadopoulos et al., 2010). Conventional MR relaxometry techniques, however, inherently require long scan times that impede the introduction in clinical practice. From a diagnostic perspective, long scan times increase the likelihood of motion artefacts, whereas from an economical perspective they reduce the throughput. In addition, long scan times cause discomfort for patients. For these reasons, MR relaxometry hasn't convinced the radiology community yet. The current project proposal aims to overcome these barriers by developing a radically new widely-applicable technological framework for accelerating MR relaxometry. At the end of this IOF SBO project, the feasibility and validity of our new approach for accelerated MR relaxometry will have been demonstrated. For final translation of the technology towards the market (and patients) we will team-up will industrial partners. Moreover, three companies (two MRI vendors and one specialized SME) already agreed to join the Industry Advisory Board and will support the project by providing early feedback. Finally, from a strategic perspective, this project bridges fundamental MR physics with applied bio-medical neuroimaging-MRI research. As such the project promotes cross-fertilization between the three Antwerp MRI-research groups (and faculties) involved. Hence, this research will enforce the mission and ambition of the University of Antwerp and its IOF consortium (Expert Group Antwerp Molecular imaging, EGAMI-image) to develop an IP portfolio and a strong translational and integrated MRI research program.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Parizel Paul
• Co-promoter: Verhoye Marleen

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Magnetic resonance imaging (MRI) is a rich and versatile imaging method able to generate images of the human body with different contrasts in a non-invasive and harmless way. Among the most popular MRI methods are diffusion MRI (dMRI) and MRI relaxometry: dMRI quantifies the mobility of water molecules, while relaxometry quantifies magnetization transfer in tissues. Acquiring both dMRI and relaxometry images of the brain is highly desirable for the following reasons: 1) both modalities provide complementary information about the brain's micro-structure and hence provide important biomarkers for brain pathologies; 2) while dMRI parameters are sensitive to brain pathologies, dMRI suffers from low specificity. Acquiring additional relaxometry data allows increasing this specificity. Unfortunately, acquiring diffusion and relaxometry parameters with sufficient resolution and precision in one imaging protocol is hardly feasible mainly due to time constraints. In this project, we will develop a paradigm shifting imaging protocol with accompanying parameter estimation framework to acquire diffusion and relaxometry parameters simultaneously. The main goal is to obtain a fingerprint image that reveals quantitative diffusion and relaxometry information in a one-stop-shop acquisition. This will not only allow to extract valuable biomarkers, but also to increase the precision and accuracy with which these parameters can be estimated in a clinically acceptable acquisition time.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Advanced composite materials (ACMs) typically contain two or more constituents, such as resin, fibers, and pores, with different physical and chemical characteristics. When combined, they produce a material with unique properties in terms of weight, strength, stiffness, or corrosion resistance. To inspect and study their 3D internal structure in a non-destructive way, the ACMs are imaged with X-rays, after which a 3D image is reconstructed from the X-ray radiographs, and further processed and analyzed in multiple sequential steps. This conventional workflow, however, suffers from inaccurate modeling and error propagation which severely limits the accuracy with which ACM parameters of interest can be estimated. In this project, we will develop a paradigm shifting approach in which the quantification of ACM parameters is substantially improved. This will be realized in a novel workflow by 1) accounting for possible deformation of the ACM during scanning, thereby reducing image reconstruction artefacts; 2) accurately modelling all constituents of the ACM (matrix, pores, and fibers); 3) directly estimating the ACM model parameters from the X-ray radiographs, thereby preventing error propagation by providing a feedback mechanism; 4) analyzing the workflow's parameter space with respect to sensitivity and stability of parameters of interest.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In this collaboration, specific reconstruction methods are being developed for extraction of quantitative information from X-ray CT images. The methods are validated on various experimental computed tomography datasets.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

We aim to create a knowledge and technology platform to promote the use of budget, low-resolution clinical scans for the production of customised lower-arm orthoses (splints) using 3D printing technology. This platform will combine statistical shape and pose models of the hand with parametric CAD design and a unique new material, resulting in an innovative splint workflow. Creating the envisioned knowledge and technology platform will steer the future of health services towards a remote, worldwide accessible patient care, offering splints at an affordable price.

Researcher(s)

• Promoter: Huysmans Toon
• Promoter: Sijbers Jan
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Across the food and manufacturing sectors, internal product defects or features related to local density differences (breakdown of tissues in fruit, cracks, badly glued seals,...) are often next to impossible to detect by conventional inline ('at-conveyor-belt') sensor technologies. These technologies provide only a surface evaluation (e.g., camera systems), a partial volume analysis (e.g., near infrared), 2D images of the product interior (e.g., X-ray radiography) or the chance of detection depends strongly on the viewing angle. Volumetric (3D) imaging can resolve such features and locate them in the product in a non-destructive way by means of X-ray computed tomography (CT). However, while conventional CT systems allow full 3D analysis, they are (1) too slow, (2) too expensive or (3) not adapted to inline applications. Today, the lack of adequate volumetric quality control in the agricultural industry results in high rejection rates (between 5 and 10% in some sectors), mostly after destructive random sampling, resulting in entire batches being removed from the supply chain. Moreover, it is also important to stress that the lack of volumetric 3D data impedes the automation in this sector. Economic stakes are therefore high. With iXCon we plan to establish a break-through in high throughput industrial quality control of products in the agricultural processing and manufacturing industry. We aim to achieve this by designing a prototype X-ray imaging system suitable for high-throughput inline imaging with the ability to perform full 3D volumetric analysis. Integrated analysis methodology will combine X-ray and sensor data (i.e. optical, laser, thermal) with prior knowledge (i.e. statistical shape or CAD models) to allow for fast 3D quality control of a level that is until now unachievable by the state-of-the-art methods.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In summary, the main goals pursued in the scope of ADAM project are: • To develop of advanced tomographic reconstruction methods for TLGI data, generating high quality reconstructions even from a limited number of projection angles and for directly estimating the material parameters of interest • To develop data fusion techniques, combinational and comparative visualization techniques enabling data overviews and detailed inspections, as well as visual analysis techniques for AC, DPC, and DFC-data of fiber-reinforced composites including bi- and multidirectional TLGI XCT data (a specimen is scanned two or more times with different orientations in order to acquire complete refraction information in all directions). A further requirement is that these techniques need to be capable of smart handling of the large TLGI XCT dataset sizes. • To evaluate the research results and to demonstrate the developed methods in a software prototype. • To disseminate the research results and acquired knowledge in order to foster the adoption of TLGI XCT inspection in industry; providing commercialization possibilities to the industrial partners and beyond.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Beenhouwer Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Advanced methods in Magnetic Resonance Imaging, such as resting state functional MRI (rfMRI) and Diffusion Tensor Imaging (DTI) will be used to study the effect of microgravity on the adaptive processes in the brain in astronauts. Preand post-flight data will be collected to elucidate changes in structural and functional brain wiring due to microgravity.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Purpose of this research community is to turn images into value by means of statistical parameter estimation. It targets to promote interdisciplinary quantitative imaging in the domain of physics, medical imaging and statistics.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Perfusion magnetic resonance imaging (MRI) is an imaging tool to assess the spatial distribution of microvascular blood flow. Many neurological disorders are accompanied by cerebral blood flow (CBF) alterations, which makes perfusion MRI indispensable in routine clinical practice. Arterial spin labeling (ASL) perfusion MRI uses magnetically labeled arterial blood water as an endogenous diffusible tracer. Tissue perfusion is measured from the signal difference between images with labeled blood and control images. Lack of ionizing radiation, complete non-invasiveness, and absolute quantification of perfusion parameters make ASL a unique perfusion imaging modality. Current ASL methods, however, suffer from problems such as noisy images and patient movement, which are inherent to the acquisition process. My project aims to develop a framework that incorporates new ASL acquisition and reconstruction methods targeting these problems simultaneously. The core of this framework revolves around super resolution reconstruction (SRR) ASL imaging which allows direct estimation of high-resolution perfusion parameters from a set of differently sampled low-resolution images. Results will yield a patient-friendly, cost-efficient and quantitative protocol that allows accurate and precise perfusion measurement at increased resolution in a clinically acceptable acquisition time, by that removing the main obstacles for ASL to become the golden standard for perfusion measurements.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: den Dekker Arjan
• Fellow: Bladt Piet

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Diffusion Weighted (DW) MRI is a unique and noninvasive method to characterise tissue microstructure, based on the random thermal motion of water molecules. Of particular interest is its potential for inferring the orientation of the coherently oriented fiber bundles within brain white matter tissue, as this opens up the possibility of investigating brain connectivity in vivo using so-called fiber-tracking algorithms. This relatively new technique is becoming a valuable diagnostic tool for a large number of neuropathological diseases. The main goals of this research topic are: - to use clinically relevant acquisitions (< 15 minutes) - to estimate fiber orientations in each voxel as accurate as possible - to assess global brain connectivity by means of tractography

Researcher(s)

• Promoter: Jeurissen Ben

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

Spherical deconvolution (SD) of diffusion-weighted MRI (DW-MRI) is a popular analysis method that allows extraction of white matter (WM) fibre orientation information in the living human brain, completely noninvasively. It can be used to track the long-range connections of the brain or serve as a tract-specific biomarker for neuronal loss in the study of neurodegenerative diseases. Recently, I proposed a new analysis method based on SD that models the presence of non-WM tissue in voxels, which was previously unaccounted for, enabling unprecedented tractography and quantification of WM. However, significant challenges remain, preventing SD from realizing its full potential: * The current approach models the signal arising from the three macroscopic tissue types. With a new approach, I want to take this to the microscopic level, taking into account the presence of axons, cell bodies and extracellular water. This will improve current neuronal fibre estimates and will introduce new quantitative measures that can be used as biomarkers in the study of neurodegenerative diseases. * Clinical scans are limited in spatial resolution due to constraints on scan time and signal-to-noise ratio (SNR). However, the very fine structures of the WM, and particularly the intricate folding patterns of the cortical surface, require high spatial resolution. I propose a new SD algorithm that can obtain high-resolution fibre information, with adequate SNR and within a practical acquisition time.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Jeurissen Ben

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

FOOTWORK's innovation is to make the digital orthotics workflow robust and automated by employing statistical foot models as prior knowledge in every phase of the workflow. Measurement phase: Statistical foot models will increase the robustness of 3D scanning wrt. noise, motion, and occlusions and allowing the use of low-cost scanners to obtain 3D accurate patient-specific foot models. Analysis phase: Statistical foot models will objectify and automate the identification of foot type and pathology based on the patient's 3D foot shape and dynamic plantar pressure. Orthotic modeling phase: Statistical foot models will enable automated and consistent design of the orthotic based on the measurements, resulting in a faster and more reliable process. In addition to these technical innovations, the consortium will also target the development of an innovative online digital workflow support platform that can be used in multiple settings, e.g. retail, orthopedics, or academic research. These innovations target a more effective and efficient digital orthotics workflow by eliminating much of the operator interaction and subjective human factors. As a result, the application of a digital orthotics workflow on a large scale becomes very attractive. Such a digital process results in faster, more reliable, and more pleasant service for the customer. It is also more economical and ecological as it involves less logistics and less material waste, keeping production in Flanders feasible. Furthermore, the opportunity to separate patient measurement from further processing allows high-tech orthopedic expert centers in Flanders to efficiently serve an international market. In that way, a larger market can be addressed more efficiently.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Huysmans Toon

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Perfusion magnetic resonance imaging (MRI) is an imaging tool to assess the spatial distribution of microvascular blood flow. Many neurological disorders are accompanied by cerebral blood flow (CBF) alterations, which makes perfusion MRI indispensable in routine clinical practice. Arterial spin labeling (ASL) perfusion MRI uses magnetically labeled arterial blood water as an endogenous diffusible tracer. Tissue perfusion is measured from the signal difference between images with labeled blood and control images. Lack of ionizing radiation, complete non-invasiveness, and absolute quantification of perfusion parameters make ASL a unique perfusion imaging modality. Current ASL methods, however, suffer from problems such as noisy images and patient movement, which are inherent to the acquisition process. My project aims to develop a framework that incorporates new ASL acquisition and reconstruction methods targeting these problems simultaneously. The core of this framework revolves around super-resolution reconstruction (SRR) ASL imaging which allows direct estimation of high-resolution perfusion parameters from a set of differently sampled low-resolution images. Results will yield a patient-friendly, cost-efficient and quantitative protocol that allows accurate and precise perfusion measurement at increased resolution in a clinically acceptable acquisition time, by that removing the main obstacles for ASL to become the golden standard for perfusion measurements.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: den Dekker Arjan
• Fellow: Bladt Piet

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

SpinOff wants to bridge the gap between modern science and high tech innovation. High school students will have the opportunity to get in contact with various high tech companies in collaboration with KHLimburg, UAntwerp, KULeuven, DSP Valley and IMEC.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

I will study a white matter model that is not restricted to coherently-oriented structures, and parameterized by several white matter tract integrity metrics which are expected to be specific biomarkers of early pathologic changes. First, I will optimize the experimental design to enable accurate and precise parameter estimation. Second, a mouse model will be used to validate and understand what the model reflect at the microstructural tissue level. Third, I will evaluate whether the parameters are markers, capable in discriminating various pathological processes of Alzheimer disease.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Van Der Linden Annemie
• Fellow: Veraart Jelle

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Our overall objective is to develop data fusion techniques that account for these spectral aspects and in this way enhance the spectral properties of a fused image or a derived product such as a land cover classification map.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

There is a wide variety of body shapes. The goal of this project is to develop a statistical shape model of the population, based on 3D scans of the exterior of the body. This virtual model is fully adjustable, both in pose as well as in body shape. The characteristics are also adjustable. The model can be used by product developers to deliver better, more comfortable, semi-custom designs.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: De Bruyne Guido
• Co-promoter: Huysmans Toon
• Co-promoter: Truijen Steven
• Fellow: Danckaers Femke

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The general purpose of this research is to model a virtual head that allows developing individualized bicycle helmets for Physical and Thermal comfort (PhyT). To this end, a CFD based thermal model will be developed as well as a shape model of the human head.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Huysmans Toon

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other hand the client. UA provides the client research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Huysmans Toon
• Co-promoter: De Bruyne Guido
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The FIAT consortium will make concrete improvements to quantitative functional imaging of tumours, which will be incorporated in clinical and preclinical application packages, clinical software modules, image analyses and ultimately routine clinical procedures.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

General purpose Our innovation goal is to realize a break-through in industrial CT image quality and to establish it as an enabling technology for the high-tech chemical, diamond and additive manufacturing industry. The main objective is the realisation of a practical, model-based iterative reconstruction platform for large industrial CT datasets that exploits prior knowledge of X-ray physics and material properties to enhance the spatial resolution.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The aim of the research is to extend our recently developed geometrical framework for spectral unmixing with recently gained insights on manifolds and the convex geometry of high-dimensional data sets. This framework will be employed for unsupervised non-linear unmixing of hyperspectral data by using geodesic distances on the data manifold.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Heylen Rob

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project is part of the BELAIR project. This project aims at the development of a Belgian super test site, where targeted Earth observation data and in situ data are collected for the Belgian and international research community.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In this project, novel computational methods are developed for optimal sampling of dMRI data that allows to either restrict the acquisition time and/or improve the accuracy of the measured diffusion profiles. Also, a general and efficient reconstruction scheme is developed to obtain high resolution dMRI images, which accounts for motion and distortion artefacts.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Cuyt Annie

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The theoretical and computer science-based expertise of Visionlab (UA) on data-driven spectral unmixing approaches and the applied biological science expertise of M3BIORES (KULeuven) on modeling and simulation-based unmixing approaches will be combined into a unique interdisciplinary framework for tackling the problem of nonlinear spectral mixtures. The shared goal is to obtain better insights into the causes of non-linear mixing and the error resulting from a linear mixing assumption, to mitigate its effects in existing unmixing applications, and to use the obtained knowledge to extract the maximum of information on the observables of the system under consideration. The developed methodology will be validated and applied to vegetation monitoring.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main goal of this project is the development of novel reconstruction and segmentation techniques for MR images which lead to improvements in the inherent trade-off between image quality and acquisition time. Prior knowledge of the grey values in the reconstructed image will be modeled in an iterative reconstruction process. The first type of grey value based prior knowledge that we will investigate is the (partial) discreteness of the grey values, analogous to the field of (partial) discrete tomography in CT. Afterwards, histogram-based prior knowledge will be introduced.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Segers Hilde

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other IBBT. UA provides IBBT research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal service agreement between UA and on the other hand BVOT. UA provides BVOT research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Huysmans Toon

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other UGent. UA shall contribute to the project under the conditions as stipulated in the present contract.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Food microstructure is defined as the organisation of food constituents at the microscale and their interaction. Most solid foods, including bakery products, fruit, vegetables and dairy foods, are microstructured. Many properties of foods which are relevant to process engineering or quality are related to their microstructure. Microstructure affects food quality attributes such as texture, but also relates to the occurrence of internal defects, as well as affecting food stability and shelf life. Examples include sponginess of bread, texture of cakes and pastry, gas and water transport properties of fruit and consistency and texture of cheeses, cream and butter. Food processing operations affects the microstructure: existing porous structures are destroyed and new ones are created. Insight in food microstructure and how it changes during processing operations is essential to produce high quality food. In particular, consumer demands for enhanced nutritional quality (composition), sensory quality (texture, internal defects) and safety (absence of foreign materials) are driving manufacturers to optimize products and processes with respect to microstructure. X-ray computed tomography (CT) enables the non-destructive visualisation and quantification of the internal structure of objects. Technological advances led to micro-CT (or CT) and nano-CT systems with nowadays a pixel resolution at or below 1 micron, while fast X-ray CT scanners have emerged in the medical field. This project with the acronym TomFood aims at - Developing novel X-ray CT instruments for inspecting food structure and food microstructure of foods at the best possible image quality and resolution balanced to processing speed and equipment cost; - Developing novel tomographic reconstruction and analysis methods for improved quantification of food structure parameters; - Using X-ray CT to improve our understanding of process-structure-property relationships through advanced mathematical models; - Develop tools for design and engineering of novel food processes and food products; - Developing affordable online food inspection equipment in food processing plants to the benefit of the food industry in Flanders. The objectives are realised by means of a multidisciplinary consortium combining food technology experts in specific application fields (dairy, fruits and vegetables, cereal based foods) with experts in X-ray physics and image processing and analysis. The objectives are translated into a program of work packages for each specific objective. The aim is to force a breakthrough in each of the domains resulting in innovations for the Flemish industry.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other Artesis. UA shall contribute to the project under the conditions as stipulated in the present contract.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Braet Johan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Vegetation and in particular green and full grown trees are extremely important in urban environments, amongst others for their beneficial effects on the reduction of air pollution. CIties are however not ideal environments for these trees and their health condition should be monitored carefully. A large scale biomonitoring of urban vegetation cannot be done manually, and remote sensing, and hyperspectral imaging in particular, announces themselves as a perfect candidate for an automated procedure. The aim of this project is to develop a framework for biomonitoring of urban vegetation from canopy spectral reflectance as this is the information that can be obtained from hyperspectral remote sensing. A data-driven approach is developed by constructing a hyperspectral reflectance library of leaf and canopy reflectance spectra. From this library, we will (i) study the relation between leaf-level and canopy-level reflectance; (ii) study the spectral distinction between healthy and unhealthy trees, and (iii) study the spectral distinction between trees growing at sites with different loads of air pollution.

Researcher(s)

• Promoter: Scheunders Paul
• Co-promoter: Samson Roeland

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other Landelijke Thuiszorg vzw. UA provides Landelijke Thuiszorg vzw research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The innovation goal of this project is to develop an algorithm to compensate respiratory related motion artifacts in chest tomosynthesis for the reconstruction of lung nodules and pulmonary lesions.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

To study functional brain networks supporting perception, action and cognition in normal and diseased human subjects and non-human primates (NHP). There will be a strong focus on anatomical, functional and effective connectivity and causality-oriented research to develop and to test biologically-relevant theoretical models for understanding brain function. The consortium will build heavily on joint expertise and collaborative experiments performed during previous phases of the IUAP program in which monkey imaging, developed by the pilot group, plays a crucial role to link human imaging studies with knowledge obtained through monkey electrophysiology.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main objective of this project is to explore and further develop the current semi-supervised and active learning techniques for the specific application of vegetation mapping. In particular, our aim is to tackle the problem of limited ground reference data by investigating the re-use of vegetation reference data. As a prototype problem, we envisage the classification of vegetation from hyperspectral images acquired at the same location on different occasions or at different locations containing similar vegetation types. The goal is then to design strategies for the re-use of reference samples obtained from one occasion or location to improve the classification at the other occasions or locations.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The goal of this project is to find the requirements for the design of a comfortable, wirelessly operating EEG headset which will be a first step towards medically approved EEG-BCI devices that can be used to communicate by patients and support clinicians in the diagnosis of patients suffering from brain disorders. New ways to model the human head will be researched using statistical analysis, different types of materials and electrodes will be compared and simulations of pressure exerted by the headset will be made in order to achieve an optimal design, which will provide an added value for all involved stakeholders. This design will then be produced and evaluated by a test panel of healthy subjects, after which it will be validated by patients suffering from ALS.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Lacko Daniël

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other hand KdG. UA provides KdG research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project proposal aims at the development of novel methods for segmentation of magnetic resonance images (MRI). Whereas conventional segmentation methods solely employ the reconstructed MRI image, we will target tomographic MRI image segmentation in which the original data from the MRI scanner is exploited. It is expected that this will lead to significantly improved segmentation accuracy.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

The SuperMRI project aims at - speeding up the acquisition process by developing new imaging sequences and sparse sampling strategies - reducing the computation time of iterative reconstruction algorithms by developing fast and generic forward and backward projectors through parallelization and distribution of the algorithms, in combination with suitable hardware architecture (GPU or FPGA). - Significantly improving the image quality by developing novel reconstruction algorithms for MRI, related to compressive sensing and discrete tomography. - developing fast tomographic image processing algorithms that exploit the available k-space data, with focus on segmentation and motion compensation.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

MESRECON aims to develope truly dual energy image reconstruction techniques. One key component of the solution is a detailed model of the acquisition and image processing. Developing such a model is a first aim of the project. The second aim is to develop dual energy reconstruction techniques based on this model and on advanced image restoration techniques.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other UZA. UA shall contribute to the project under the conditions as stipulated in the present contract.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Computed Tomography (CT) is a powerful, nondestructive technique for producing 2-D and 3-D cross-sectional images of an object from X-ray images. Conventional CT requires the object to be inserted into the CT scanner and that many X-ray images are taken, prior to reconstruction of the 3D image. However, there are many situations in which these requirements are not met: If the object is a fixed part of an object that is too large to fit in the scanner object; if moving the object is dangerous (e.g., explosives), if moving the object would disrupt or pollute the context (e.g., crime scene), if the object cannot easily be transported for X-ray scanning (e.g., horse with broken bone), or if the object is too valuable to be removed (e.g., cultural heritage). To cope with such situations, in situ X-ray scanning is required. Portable X-ray devices, such as a hand-held X-ray camera or a robot system, are available on the market, which are, however, intended to acquire only a single or a series of X-ray images. It is currently not possible to produce a 3D reconstruction of the object from these X-ray images. This is because: 1. in contrast to common CT-scanners, the exact position and orientation of the X-ray source/detector system with respect to the object is unknown. 2. if a hand-held camera or robot system is employed, the scanning process is time consuming, which limits the number of X-ray images to be acquired for tomography. Current CT reconstruction algorithms require a large number of X-ray images to obtain accurate results. 3. it may not be possible to acquire projection images from all angles. Both 2 and 3 result in a highly underdetermined inverse problem. This project aims at the development of robust, efficient reconstruction methods for in situ X-ray scanning & tomography. These methods will not require accurate prior knowledge of the scanning geometry, and will be tailored for achieving maximal reconstruction quality from a small number of projection images. To this end, the following computational strategies will be combined: 1. Automatic parameter estimation based on consistency maximization of the simulated X-ray images with respect to the measured X-ray images will allow the reconstruction algorithm to deal with unknown geometrical parameters. 2. New algorithms will be developed for efficient optimization of the high dimensional search space (including the unknown object volume and the position/orientation of the acquired X-ray images) by exploiting the linearity of certain reconstruction algorithms and exploring multi-resolution approaches for gradual refinement of parameter estimates. 3. Compressive sensing and discrete tomography will be incorporated within the parameter estimation framework to allow for accurate image reconstruction from few projections and deal effectively with a limited angular range. 4. State-of-the-art GPU computing techniques, based on recent advances in a current research project on accelerating tomography algorithms, will be employed to effectively deal with the high computational requirements. A successful project will open up numerous applications in various fields such as security, nondestructive testing, dental imaging, veterinary imaging, or cultural heritage.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other hand KHL. UA provides KHL research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Frans Renaat

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The goal of CHAMELEON is to design the next generation vision system that exploits the benefits of spectral information. The mission and novelty of CHAMELEON is: To enable and demonstrate flexible, but domain-specific hyperspectral imaging systems for relevant industrial applications. More specifically, Chameleon will promote direct usage of hyperspectral imaging by the industry by providing flexibility in terms of: • System exploration and simulation tools • Novel camera architectures, including parameterizable templates • Novel hyperspectral image processing methods and techniques • Mapping, data handling and interpretation strategies for real-time operation • Novel measurement strategies

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Sinusoidal analysis is an important technique used to analyse audio. However, when analysing polyphonic audio, the harmonic components of different notes tend to overlap and the resulting system matrix cannot be solved. Using additional information, mainly from a music score, should enable us to start from a better initial guess of the model, and determine the correct number of parameters beforehand. This should result in a much more robust and accurate analysis. The first part of this project is concentrated around the mathematical an technical problems involved, and will also consist of the development of the necessary software tools. The second part of the project focuses on applications, more specifically on source separation or elimination and on individual note editing in a polyphonic context.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Ganseman Joachim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a research agreement between the UA and on the onther hand IWT. UA provides IWT research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Veraart Jelle

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

An extensive study on the use of mathematical morphology for the processing and analysis of multispectral and hyperspectral image data is carried out, taking into account both the spectral and spatial property of the data. State-of-the-art techniques are enhanced and new methods are developed. As application we investigate the use of colour and multidimensional image processing in remote sensing, i.e. the classification of hyperspectral images using mathematical morphological profiles for vegetation mapping and habitat monitoring.

Researcher(s)

• Promoter: De Witte Valérie

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The project will in general investigate and develop a number of innovative imaging, image reconstruction and analysis techniques specifically for lung diseases, which can become market ready in a few years.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The increasing amount of image data available from airborne and satellite sensors demands new strategies to overcome the trade-off in spectral and spatial resolution. in this project, two different approaches to enhance the spatial information in hyperspectral image are tackled: state-of-the-art data fusion and advanced unmixing techniques.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other hand Handi-Move. UA provides Handi-Move research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The purpose of this research is in the first place to model the Adjacency effect (i.e. to distinguish the different components of the radiance) and to develop correction algorithms that can be applied to high resolution data sets. These can be hyperspectral as well as multispectral image data. At the start, the Adjacency effect will be studied on the separation between water and land, but the general purpose is to develop a general analysis method, leading to generic correction algorithms.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Geens Bert

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Concretely, we will develop a DTI-atlas for the premature brain (0 to 4 year) by combining non-affine registration techniques and new denoising algorithms for DWI-data as well as accounting for different growing speeds of the premature brain. Nowadays, DWI is also seen as one of the fastest growing modalities for preterm brain analysis.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The SuperCT project aims at 1) reducing the computation time of several tomographic image reconstruction algorithms. In particular, we will focus on developing fast and flexible iterative reconstruction techniques for X-ray CT, SPECT, and Electron Tomography. Most of the speedups will be obtained by parallelization and distribution of the algorithms, in combination with suitable hardware architecture (GPU). 2) developing and implementing fast tomographic image processing algorithms that exploit the available projection data. In particular, we will focus on segmentation, beam hardening correction, visualization, and alignment.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project concerns the inclusion of the ISOMAP non-linear reduction algorithm into two existing hyperspectral image analysis techniques: texture classification by wavelet distribution based geodesic distance calculations, and endmember extraction for unmixing purposes. The former technique is recently developed at the Visionlab, where the latter will provide an entirely new approach to an existing problem.

Researcher(s)

• Promoter: Heylen Rob

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Epilepsy is a frequently observed neurological disorder that is characterized by repeated epileptic seizures. These seizures are characterized by a sudden and unexpected change of the behavior and/or the consciousness of the patient as a result of excessive and uncontrolled electrical activity in a well-defined region in the cerebral cortex (the so-called epileptogenic zone). By placing electrodes using a standardized setup on the scalp surface, one can measure the electrical or magnetic fields generated by brain activity during an epileptic event. A recording of electrodes in function of time is called an electroencephalogram (EEG) or magnetencephalogram (MEG), respectively. EEG/MEG source analysis determines the origin of brain activity based on the EEG/MEG due to epileptic events and consists of two subproblems: First, by solving the forward problem one obtains the electrode potentials due to a given set of sources, which are characterized by the source parameters. Second, by solving the inverse problem the source parameters are estimated given a set of measured electrode potentials. These imaging techniques measure the generated brain activity with a high time resolution (milliseconds) but have a low spatial resolution. In the past decade, a vast number of medical imaging modalities have been used in the clinical practice. Magnetic Resonance Imaging (MRI), Computed Tomography (CT) and Diffusion Weighted MRI (DW-MRI) are techniques that reveal the anatomical structure of the brain. Functional imaging, such as Single Photon Emission Computed Tomography (SPECT) and functional MRI (fMRI), image changes in the blood flow. These techniques have a high spatial (approx. 1 mm for MRI, 7 mm for SPECT, 3 mm for fMRI) resolution, but a very low temporal resolution as the duration of the scan is in the order of minutes. Novel treatment options have emerged from the experimental field, which involve the stimulation of a specific brain region, deep brain stimulation (DBS), or vagal nerve, vagal nerve stimulation (VNS). However, the mechanism of action of these procedures is unknown and the efficacy of the treatment can be improved. Within the iMIND project we want to develop a software platform that can: 1. Gather the necessary information for the determination of the origin of epileptic seizures and the quantification of the mechanism of action of novel neuromodulatory treatment. 2. Coregistration of the different acquired images in order to visualize and analyze them in the same frame of reference. Furthermore we want to improve the accuracy of the EEG/MEG source analysis by incorporating anatomical and functional information obtained from medical imaging 3. Visualize the images in a comprehensive and user-friendly way. We also want to visualize the time information, provided by the improved EEG/MEG source analysis procedure. 4. Be used for the accurate determination of the epileptogenic zone. In this case we want to determine the added value of the software platform on a small population in the determination of the epileptic onset zone. 5. Be used in an experimental setting by quantitatively measuring the effect of different parameters of novel neuromodulatory treatment (DBS and/or VNS). In this case, the added value is determined by using the software platform for comparing functional images obtained during stimulation of small animals and correlating them with anatomical images.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Medical imaging is getting more and more complex. Unfortunately the medical imaging community today by far is not optimally making use of color and multi-dimensional information. This research project will significantly improve this situation. In the technology working packages we will develop platform technology to better handle multi-dimensional and color data. The basic technology that will be developed in this project covers the entire imaging chain starting with acquisition device over image processing to visualization and finally (clinical) validation and standards. This technology will be use as a basis in other working packages and applied to specific clinical applications.

Researcher(s)

• Promoter: Scheunders Paul
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a formal research agreement between UA and on the other hand EU. UA provides EU research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Sinusoidal analysis is an important technique used to analyse audio. However, when analysing polyphonic audio, the harmonic components of different notes tend to overlap and the resulting system matrix cannot be solved. Using additional information, mainly from a music score, should enable us to start from a better initial guess of the model, and determine the correct number of parameters beforehand. This should result in a much more robust and accurate analysis. The first part of this project is concentrated around the mathematical an technical problems involved, and will also consist of the development of the necessary software tools. The second part of the project focuses on applications, more specifically on source separation or elimination and on individual note editing in a polyphonic context.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Ganseman Joachim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Veraart Jelle

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Van Hecke Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The project proposes to research the following ¿ Definition of a vision beyond Web2.0/Telco2.0/Enterprise2.0 starting from an in depth analysis of market and technology trends. ¿ Implementation of the vision under form of the definition of a WTEPlus framework. ¿ Research on a number of key functional elements of the framework: workflow and service assurance ¿ Development of a number of Proof of Concepts in the three application domains.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The goal of this project is to develop new statistical models for spatial classification procedures for multispectral and hyperspectral remote sensing data, in which spectral, polarimetric and spatial information is treated simultaneously. The methodology is based on the following principles: 1. the use of multiresolution techniques 2. the use of Bayesian principles 3.the use and the development of new multivariate Markov Random Field models

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project represents a research agreement between the UA and on the onther hand IWT. UA provides IWT research results mentioned in the title of the project under the conditions as stipulated in this contract.

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

X-ray tomography is a powerful technique for recovering the three-dimensional (3D) structure of an object in a nondestructive manner. Since its introduction in the 1970's, tomographic scanners have become standard equipment in medical facilities. More recently, tomographic scanners have been developed that can be used for non-clinical applications in biomedical and materials imaging. Unfortunately, tomographic reconstructions often contain artifacts. These artifacts can be induced by the hardware of the scanner, the reconstruction algorithm or the sample itself. All these types of artifacts are being addressed in the TIRPA project aiming at improving the various steps between acquisition of the projection data and segmentation of the reconstruction, to reduce the errors introduced in each of the steps and, thereby, improve the quality of the reconstruction. An example of artifacts due to hardware is the problem of ring artifacts coming from inaccuracies in the scanner's source or detector. This problem will be tackled using a new method of sinogram processing. Secondly, the presence of high density structures in the object such as metal leads to streaking artifacts induced by the standard reconstruction algorithm. Different interesting and innovative reconstruction algorithms are implemented to overcome these problems. Third, if the scanned object is larger than the field of view of the measured projections, truncation artifacts arise. In order to study the influence of the different parameters on the image quality, a cone-beam computed tomography simulator will be built. The algorithms that automatically search for the optimal parameter setting for a given imaging problem are also investigated making the simulator a versatile tool for CT-users. The simulator will be further extended with a metal artifact suppression algorithm. Thereby, we will adopt a recently proposed, iterative reconstruction algorithm that first detects the metal structures and next, using a prior on the local intensity distribution, performs an iterative reconstruction. Furthermore, a new method will be developed that will combine continuous with discrete tomography to reduce metal artifacts. Finally, a novel ROI reconstruction technique for truncation artifact reduction will be investigated in which the shape and structure outside the field of view are approximated by a uniform ellipse. Thereby, the parameters of the ellipse (density, axes, position and orientation) will be optimized using specific consistency conditions. Aforementioned CT techniques are used in vivo for anatomical imaging. Computed tomography can however also be used for functional imaging as described in the last part of this project where CT is used for (pre)clinical molecular imaging. In this case CT is abbreviated as SPECT (Single Photon Emission Computed Tomography). Current SPECT hard- and software delivers suboptimal specificity due to a lack of sensitivity, resolution or by introduction of artefacts. Current reconstruction methods for SPECT are not patient specific and accordingly personalized treatment is not possible.. It is the purpose to innovate iterative computed tomography reconstruction techniques for human (brain) imaging through the use of the statistical Monte Carlo model to correct for errors in the imaging process.. Such a quantitative image reconstruction enhances diagnosis and ultimately patient comfort. Also the use of CT for molecular imaging of small animals (microSPECT) will be investigated: we will perform gated and dynamic small animal experiments with various isotopes using novel radiopharmaceuticals with high end microSPECT hardware. Furthermore, the combination (fusion) of the SPECT images with the micro-CT data which is of utmost interest in the field of small animal imaging where the micro-SPECT and micro-CT are relatively new techniques. Accordingly, this project is not only focused on new techniques for problem solving in general but it is applied on specific application areas as well. The differences in these areas prove the strength of the ideas. The specific application fields are chosen from diamonds to foam material over pre-clinical studies to human clinical applications.

Researcher(s)

• Promoter: Sijbers Jan

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

This project focuses on reducing the scan and reconstruction time for various forms of Computer Tomography. Iterative reconstruction algorithms require fewer projections (= shorter scan time) to obtain an accurate reconstruction. However, the computation time of such methods is currently too long for practical use. By developing algorithms for the GPU (GPU computing), the reconstruction time will be strongly reduced.

Researcher(s)

• Promoter: Batenburg Joost

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Diffusion Tensor Magnetic Resonance Imaging (DT-MRI) is a recently developed technique who permits to study the architecture of white brainmatter (WM) in vivo and in an non-invasive way. DT-MRI is based on the Brownian movement of H2O-molecules in biological tissue and makes it possible to determine the anisotropic diffusion of these molecules . This anisotropic diffusion can be related to aligned microstructures, like WM brain fibres, which has a great value in biomedical applications. Since a large amount of data is needed for this technique, it is desirable to use fast imaging sequences. However, these kind of sequences introduce specific artefacts in the images which degrade the quality of the DT-measurements. For this reason, several strategies will be used to upgrade this quality. The present acquisition standard for fast DTI, Echo Planar Imaging (EPI), is prone to severe susceptibility artefacts which introduce geometric distortions in the images. These artefacts are more explicit when working at higher field strengths (here: 7 Tesla and 9.4 Tesla). By using an adapted EPI-sequence, it is possible to measure the local susceptibility artefacts and to correct for distortions. Another strategy that will be used is to combine DTI with Fast Spin Echo (FSE). This technique should be less sensitive to susceptibility artefacts. A recent approach, in which multiple receivers are used (Parallel Imaging) will be used to reduce artefacts in DT-MRI.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Verhoye Marleen
• Fellow: Pintjens Wouter

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project will concentrate on a particular image processing and image analysis problems. This problem is about creating and analyzing large biomedical image datasets. When the main goal is analysis, or CT reconstrcution, there is a need for automatic construction of a so called mosaic images, aligning and combining all images, creating a large FOV (field-of-view) image.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The focus of this project is signal processing in functional MRI. Functional brain imaging offers a way to image the specific brain areas that are active during a specific action. Brain activation is present in the MRI images due to the BOLD (Blood Oxygen Level Depended) response. Since the BOLD response signal is weak compared to the noise level accurate modeling of the response as well as the disturbances will improve brain activation detection. Within this project the time and space correlations of fMRI signal are modeled. The model will be used to derive estimators for task depended brain activation. We seek to improve the current standard processing of fMRI datasets in the following places. The time correlation is currently often modeled by an AR(1) process (autoregressive process with 1 degree of freedom). There are indications that this is not always an accurate description of the disturbances, therefore methods are developed that automatically select the best AR order and model from the data to optimally model the temporal noise structure. Usually for activation detection a linear regression is performed with the stimulus paradigm convolved by a HRF (hemodynamic response function) as regression vector. However it is known that the HRF is not constant among brain regions and subjects. Therefore we will investigate ways to estimate the HRF from the measured data. To decide which part of the brain are active statistical test have to be performed. In order to make these tests accurate the noise level of the images often is needed. This noise level can be robustly estimated from background area present in the images. Traditionally the maximum of the background mode of the histogram is used as an estimator for the noise level. In this project we develop a Maximum Likelihood method which can robustly determine the noise variance from the histogram of the background mode of the image.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Poot Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Project description Transmission tomography plays an important role in medical and industrial applications. It allows the reconstruction of objects using a large number of their projections, gathered from different angles. The currently most used algorithms are types of Filtered Back-Projection (FBP), such as the Feldkamp-algorithm. Although they can be implemented efficiently, this algorithm has large drawbacks. First, a large number of projections is required to guarantee an accurate reconstruction. Secondly, FBP is very susceptible to noise and thirdly, it does not allow for using prior knowledge during a reconstruction. This project will use iterative reconstruction algorithms, which don't suffer from the previously mentioned draw-backs. The following problems will be solved: Long scanning time: because the number of required projections is high, this also causes a long scanning process. A large number of application require a short scanning time, such as: scanning of a patient, to increase patient comfort. Reducing the number of scans will also reduce the received scanning dose. It will also allow for scanning dynamical processes, because it will decrease the exposure time and thereby reduce motion artifacts. Industrial applications require shorter scanning times to speed up checks. Long reconstruction time: when using algebraic reconstruction methods, the resulting reconstruction will be more accurate, but the reconstruction time will be much longer. This is caused by the increased number of calculations required to do a reconstruction. To solve this, we will introduce new high-performance hardware: Graphical Processing Units. GPUs are responsible for the rendering of a 3D scene on a computer monitor. Their extreme parallel nature gives calculation speeds unheard of on CPUs. Most computations done during a reconstruction using an algebraic reconstruction algorithm can be parallelized and therefore use the power of the GPU. Large amount of required memory: when reconstructing a large 3D volume, the required memory space is gigantic, too large to be stored on one GPU. But because all GPUs are required to access all the memory during a reconstruction, ways will have to be developed to allow multiple GPUs to work on one dataset. Simple solutions are naïve partitioning and caching schemes. More elaborate solutions will be considered to reduce the reconstruction time, while keeping the reconstruction quality. Long implementation time when testing new iterative reconstruction algorithms: when considering iterative reconstruction algorithms at a higher level, one will notice their overall similarity. They all feature a projection-step, an error-calculation-step and a back-projection-step. Yet every programmer interested in applying an algebraic reconstruction algorithm has to write a completely new program implementing the functionality. It becomes even more work when one wants to implement it on a GPU. Then also the ns and outs of that hardware needs to be understood. A code generator will be implemented to produce code that will be composed of the required sub-units. This way different types of projection and back-projection steps can be chosen and put together. This combined with a static study of the code (to allow for a more efficient use of resources)

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: van der Maar Sander

Research team(s)

• Vision lab

Project website

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Batenburg Joost

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Traditional image processing and analysis techniques have been developed for scalar images. However, with the evolution of imaging technology, often multivalued images are produced. In contrast to scalar images, where every pixel contains a single gray value, the pixels of these images contain vectorial information. Examples are color images, multispectral images and multimodal biomedical images. Image segmentation is an essential process for most subsequent image analysis tasks, because low-order pixel based information is converted into region based higher-order information, which allows to analyze the information more efficient. Image restoration is also an important preprocessing step, because homogeneous images, where the noise was removed, are easier to process with a segmentation algorithm. The goal of this project is to develop a new technique for segmentation of multivalued images. The algorithm will contain a preprocessing procedure to reduce the noise. The proposed technique will be based on the combination of the following three principles: ¿Interband approach: By processing the information of all bands simultaneously, both the spatial and the spectral information is used for the restoration and segmentation. This way, all the available information is fully exploited. ¿Multiresolution approach: Multiresolution techniques such as the wavelet transform decompose an image in several resolution scales. Restoration is useful in this representation, because the noise amplitude typically decreases at the lower resolution scales. For segmentation, a coarse segmentation on lower resolution scales can be used to obtain a finer segmentation on higher resolution scales, with a hierarchical procedure. ¿Model based approach: An a-priori model is assumed for the probability density function (pdf) of the pixels. The parameters of the model are estimated by means of the available images and this data can be used for restoration and segmentation. The second part of this project consist of the validation of the proposed techniques. This validation will be performed on two different research fields with realistic applications: ¿Multispectral images (earth observation). By means of cooperation of the Visionlab with the Teledetectie and atmospheric processes (TAP) department of the VITO, there are multispectrale and hyperspectrale datasets available with applications in forestry, vegetation monitoring, ground pollution with heavy metals, etc.... ¿Multimodal biomedical images. The visionlab has a cooperation with the research group 'Bio-imaging Lab' of the University of Antwerp, where magnetic imaging (MRI) is used for biomedical research.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Driesen Jef

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

DTI is a unique technique that provides in vivo and non-invasive information regarding the organisation and structural integrity of tissues. It is commonly used to study all kinds of brain diseases (MS, Parkinson, etc). An important issue is the early quantitative detection of brain abnormalities. However, that is only possible when non-affine registration techniques are avaliable that can deal with multi-valued data from DTI images. The goal of this project is to develop such registration techniques.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Van Hecke Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project aims to develop a operation-oriented methodology to map, to monitor and to evaluate habitat characteristics, vegetation types and gradients in order to determine nature quality. While the proposed methodologies are generic enough to support any classification problem with remote sensing and applicable in many domains, our focus is a transferable platform for habitat status reporting. A major application, and for the test and validations of this project, is the detailed mapping of ecotopes that can assess the conservation status of Natura 2000 habitats. . In remote sensing, we most often have to deal with sensor limitations. No single sensor combines the optimal spectral, spatial and temporal resolutions. Mapping of ecotopes would be difficult without hyperspectral data. Spaceborne hyperspectral data, which give good large coverage, is always coarse in resolution. To improve spatial resolutions of spaceborne hyperspectral imagery, we propose the use of superresolution image reconstruction techniques which utilize the complementary information in multi-angle overlapping imagery. Conventional classification methods do not address spatial and structural dimensions adequately for habitat reporting. One objective of this project is to make optimal use of available sensors. Spectral information will be extended to include multi-scale information, spatial features (classification of segmented regions, textural/contextual features), and post-processing of classification results (clustering, rule-based learning). Structural analysis will target the composition of vegetation types that can be used for diagnosis of habitat quality as well as future change modeling. The innovative of this proposal is the combination of all these techniques to fully exploit the available data. Rather than just improving state of the art spatial features and superresolution techniques, this project will assess the implication of the improved features on the classification performance. Furthermore, the classification framework will be enhanced by introducing ensemble classifiers. The objective is to investigate the operational potential of ensemble classifiers in terms of stability, accuracy, ease of use, and computing costs. The developed algorithms and methodologies aimed for habitat status reporting will be integrated and tested by a very specific application: assessing the conservation status of Natura 2000 habitats in Special Areas for Conservation' (SACs). While this proposal is composed of heavy methodological elements, user's requirements will provide all the guidelines for developments and integrations as well as defining expected results. INBO and Alterra, both specifically involved in the implementation of the EU directive to report on Natura 2000 habitats, and the mapping of vegetation and habitat types in general, are ideal partners in the domain of habitat reporting. This cross-disciplinary proposal will be first to address operational issues with habitat status reporting using novel and advanced remote sensing methodologies. The end results will have significant implications on future methodologies for biodiversity and habitat monitoring. The outcome from the technical aspects will be beneficial to some core scientific issues and generic developments of remote sensing methods .

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: Verdonk Brigitte

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The project is focused on the problem of accurate correction of geometrical distortions in a 2D X-ray CCD detector. Proper phantoms and numerical methods are to be developed to eliminate hardware problems that occur due to the unavoidable errors in the X-ray detector manufacturing process (errors in glass waveguide production). Results can be applied to any 2D detector where geometrical distortions occur.

Researcher(s)

• Promoter: Postnov Andreï

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Region of interest (ROI) cone-beam tomography has become a hot topic in the continuous quest for reducing the amount of radiation and achieving a higher resolution of the object through geometric magnification. This magnification is achieved by moving the object closer towards the object so that the ROI fully covers the field of view. A disadvantage of this method is truncation of other parts of the object, while in theory all information about the object is needed for ideal reconstruction. In this project, two reconstruction algorithms are examined that, in combination with existing methods, may result in a significant improvement of the reconstruction quality. In the research group Visionlab, a new algorithm for ROI reconstruction is developed and is already implemented for a parallel geometry. The algorithm reduces the effect of the attenuation of the X-rays outside the ROI using a Gaussian window function. Preliminary results show that in case the parameters are adjusted optimally, good results are achieved. The goal of this project is now to further examine the algorithm and to improve it where its possible. The influence of noise will be examined and the method will be validated for real data and different acquisition geometries. The second algorithm, the universal reconstruction algorithm (URA) calculates a reconstruction of a general acquisition, for any geometry. The algorithm builds up the reconstruction image ray by ray in the frequency space. This is followed by an interpolation in order to fill up a regular lattice. After normalization and correction for unequal sampling, we get the image by performing an inverse Fourier transform. The URA will be analytically and numerically examined for both general acquisition problems (such as helical cone beam) and ROI problems.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Van Gompel Gert

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The focus of this project is signal processing in functional MRI. Functional brain imaging offers a way to image the specific brain areas that are active during a specific action. Brain activation is present in the MRI images due to the BOLD (Blood Oxygen Level Depended) response. Since the BOLD response signal is weak compared to the noise level accurate modeling of the response as well as the disturbances will improve brain activation detection. Within this project the time and space correlations of fMRI signal are modeled. The model will be used to derive estimators for task depended brain activation. We seek to improve the current standard processing of fMRI datasets in the following places. The time correlation is currently often modeled by an AR(1) process (autoregressive process with 1 degree of freedom). There are indications that this is not always an accurate description of the disturbances, therefore methods are developed that automatically select the best AR order and model from the data to optimally model the temporal noise structure. Usually for activation detection a linear regression is performed with the stimulus paradigm convolved by a HRF (hemodynamic response function) as regression vector. However it is known that the HRF is not constant among brain regions and subjects. Therefore we will investigate ways to estimate the HRF from the measured data. To decide which part of the brain are active statistical test have to be performed. In order to make these tests accurate the noise level of the images often is needed. This noise level can be robustly estimated from background area present in the images. Traditionally the maximum of the background mode of the histogram is used as an estimator for the noise level. In this project we develop a Maximum Likelihood method which can robustly determine the noise variance from the histogram of the background mode of the image.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Poot Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Industrial vision applications in which colour evaluation forms the key problem, is a separate and rather difficult part in vision technology. An industrial accepted calibration method will help in solving a lot of practical problems than and will give progress to the use of colour vision applications. The Skr-matrix method, for which patent PCT/EP2005/003889 is taken by the Antwerp Innovation Centre and which is developed at the Lab for Industrial Vision, forms a powerful Colour Handling Method which can be used in all kinds image handling industries. Especially, we think about digital camera's, scanners, copiers, screens, IR-camera's...

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Fellow: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In this project, improved statistical tests will be developed voor functional magnetic resonance imaging (FMRI) data analysis. Thereby, parametric tests, based on generalized likelihood ratio tests, as well as non-parametric tests (such as clustering) will be investigated. Finally, advanced methods for visualisation of fMRI detection results on high resolution MR images will be developed.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: D'haes Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Fellow: D'haes Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Traditional image processing and analysis techniques have been developed for scalar images. However, with the evolution of imaging technology, often multivalued images are produced. In contrast to scalar images, where every pixel contains a single gray value, the pixels of these images contain vectorial information. Examples are color images, multispectral images and multimodal biomedical images. Image segmentation is an essential process for most subsequent image analysis tasks, because low-order pixel based information is converted into region based higher-order information, which allows to analyze the information more efficient. Image restoration is also an important preprocessing step, because homogeneous images, where the noise was removed, are easier to process with a segmentation algorithm. The goal of this project is to develop a new technique for segmentation of multivalued images. The algorithm will contain a preprocessing procedure to reduce the noise. The proposed technique will be based on the combination of the following three principles: ¿Interband approach: By processing the information of all bands simultaneously, both the spatial and the spectral information is used for the restoration and segmentation. This way, all the available information is fully exploited. ¿Multiresolution approach: Multiresolution techniques such as the wavelet transform decompose an image in several resolution scales. Restoration is useful in this representation, because the noise amplitude typically decreases at the lower resolution scales. For segmentation, a coarse segmentation on lower resolution scales can be used to obtain a finer segmentation on higher resolution scales, with a hierarchical procedure. ¿Model based approach: An a-priori model is assumed for the probability density function (pdf) of the pixels. The parameters of the model are estimated by means of the available images and this data can be used for restoration and segmentation. The second part of this project consist of the validation of the proposed techniques. This validation will be performed on two different research fields with realistic applications: ¿Multispectral images (earth observation). By means of cooperation of the Visionlab with the Teledetectie and atmospheric processes (TAP) department of the VITO, there are multispectrale and hyperspectrale datasets available with applications in forestry, vegetation monitoring, ground pollution with heavy metals, etc.... ¿Multimodal biomedical images. The visionlab has a cooperation with the research group 'Bio-imaging Lab' of the University of Antwerp, where magnetic imaging (MRI) is used for biomedical research.

Researcher(s)

• Promoter: Scheunders Paul
• Fellow: Driesen Jef

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Gelieve aan te vullen a.u.b.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Van Der Linden Annemie
• Fellow: Leemans Alexander

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

In many industrial vision settings one makes use of an image sequence. These multiple view systems give rise to specific questions on: -the required geometric calibration in order to perform measurements . -corresponding points/line segments/regions , -the relative camera motion between successive views , -the 3D-reconstruction of objects , The literature on computer vision offers an extensive treatment of these topics, however often formulated in advanced algebraic and geometric terms. It is our goal to translate the theory into ad hoc algorithms, directly applicable to a given industrial environment.

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: Peremans Herbert

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The significant decline in the incidence of dental caries that is observed in the last two decades [Sawle et al., 1988] [Kalsbeek et al., 1998], also seems to be accompanied by a distinctive change in lesion morphology. In particular, the phenomenon of slowly progressing dentinal lesions, obscured by apparently intact enamel [Kidd et al., 1992] [Poorterman et al.,1999], renders the established diagnostic procedures of visual examination and intra-oral radiography, ineffective [Poorterman et al.,2000]. However, reliable detection at an early stage of the carious progression remains crucial to allow for an appropriate preventive or minimal invasive treatment. Extraordinary sensitivity scores, however, have been achieved with diagnostic methods based on the premise that demineralized lesions show a significant higher electrical conductivity than healthy tissue, due to the phenomenon of increased pore volume. Even more impressive results were attaind by means of spectroscopic conductance measurements. Clinical application of these electrical conductivity methods, unfortunately seems to be hampered by a lack of reproducibility. Variations in conductivity readings makes it very difficult, if not impossible, to define generally applicable diagnostic thresholds for discriminating among carious involvement. It may be hypothesized, that this problem is related to the very nature of the 2-point measurement configuration underlying all of the experimental methods, described in the literature as today. This study instead, aims to develop and evaluate a new method of spectroscopic Electrical Impedance Tomography (EIT), capable of reconstructing cross-sectional maps of the coronal tooth structure, depicting site-specific electrical impedance spectra. The resulting method promises to improve upon existing electrical caries detection methods, both in terms of its improved cross-sectional sampling strategy, dispensing with the need to rely on visual surface indications to determine appropriate measurement sites, and its immunity with regard to the natural variability in electrical conductance between individual teeth. Tomographic representation of the measurement results will allow the diagnostic interpretation to proceed on the basis of relative changes in tissue impedance among different spatial locations, instead of being dependent on a single quantitative interpretation. The suggested spectroscopic extension of the established method of EIT , together with the particularly difficult measurement conditions created by the insulating properties of the outer enamel layer, will pose some very specific problems as to the mathematical demands on the reconstruction algorithms. Therefore, the study and evaluation of non-linear reconstruction algorithms for the inverse conductivity problem of EIT, based on concepts drawn from numerical algebra, will constitute the major contribution in this research proposal. Despite the fact that evaluation of the algorithmic developments will be limited to numerical simulation and in vitro experiments, design of the newly developed method will be such, that its essential properties can be expected to transfer straightforwardly to the clinical setting. To gain a more fundamental understanding of the electrical conduction phenomena taking place in dental structures, a first, morphological accurate, numerical simulation model of a whole tooth will be built. The detailed morphological input data for this model will be derived from micro-Computed Tomography (micro-CT) scans, acquired with a microfocal desktop-sized measurement system. Cross-sectional images acquired in this way, will be processed by an automatic segmentation procedure, before being converted into a 3D Finite Element (FE) mesh. Such a mesh description allows to build up the actual volumetric simulation model.

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: De Clerck Nora

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The main goal of the project is the development/improvement of advanced forms of tomography and its application in practice. On the one hand attention will be paid to the development of tomography at the nanoscopic level by means of EFTEM while on the other hand element-specific X-ray fluorescence tomography with micrometer resolution will be optimised and quantitatively calibrated. An important area of application is the quantitative three-dimensional measurement of various phases in (human and animal) bone.

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: D'Haese Patrick
• Co-promoter: Janssens Koen

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The topic of this project is a nalysis and segmentation of multispectral images with applications in remote sensing . We will develop multispectral versions of low-level image processing operators, multispectral edge detectors and techniques for multispectral texture characterisation. We will vaildate the techniques in remote-sensing applications.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Region of interest (ROI) cone-beam tomography has become a hot topic in the continuous quest for reducing the amount of radiation and achieving a higher resolution of the object through geometric magnification. This magnification is achieved by moving the object closer towards the object so that the ROI fully covers the field of view. A disadvantage of this method is truncation of other parts of the object, while in theory all information about the object is needed for ideal reconstruction. In this project, two reconstruction algorithms are examined that, in combination with existing methods, may result in a significant improvement of the reconstruction quality. In the research group Visionlab, a new algorithm for ROI reconstruction is developed and is already implemented for a parallel geometry. The algorithm reduces the effect of the attenuation of the X-rays outside the ROI using a Gaussian window function. Preliminary results show that in case the parameters are adjusted optimally, good results are achieved. The goal of this project is now to further examine the algorithm and to improve it where its possible. The influence of noise will be examined and the method will be validated for real data and different acquisition geometries. The second algorithm, the universal reconstruction algorithm (URA) calculates a reconstruction of a general acquisition, for any geometry. The algorithm builds up the reconstruction image ray by ray in the frequency space. This is followed by an interpolation in order to fill up a regular lattice. After normalization and correction for unequal sampling, we get the image by performing an inverse Fourier transform. The URA will be analytically and numerically examined for both general acquisition problems (such as helical cone beam) and ROI problems.

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Van Gompel Gert

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Hyperwave belongs to the research domain of hyperspectral image processing. Hyperspectral sensors have a high spectral resolution of a few nanometers. Opposite to multispectral sensors, they can register small differences in spectral signature. The disadvantage is that they produce an enormous amount of data. It is impossible to process the data without reduction. The challenge is to extract the useful information from the raw data and to ignore redundant data. A previous project, Hypercrunch (SR/00/05) was a firs onset to this. However, the application, detection of stress in orchard, was very specific and extremely ambitious. With this project, we aim at further developing the algorithms and using them for totally different applications. The final goal is to build the algorithms into processing chains of hyperspectral data, in view of the operational exploitation of the hyperspectral APEX instrument, that start in 2005.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The significant decline in the incidence of dental caries that is observed in the last two decades [Sawle et al., 1988] [Kalsbeek et al., 1998], also seems to be accompanied by a distinctive change in lesion morphology. In particular, the phenomenon of slowly progressing dentinal lesions, obscured by apparently intact enamel [Kidd et al., 1992] [Poorterman et al.,1999], renders the established diagnostic procedures of visual examination and intra-oral radiography, ineffective [Poorterman et al.,2000]. However, reliable detection at an early stage of the carious progression remains crucial to allow for an appropriate preventive or minimal invasive treatment. Extraordinary sensitivity scores, however, have been achieved with diagnostic methods based on the premise that demineralized lesions show a significant higher electrical conductivity than healthy tissue, due to the phenomenon of increased pore volume. Even more impressive results were attaind by means of spectroscopic conductance measurements. Clinical application of these electrical conductivity methods, unfortunately seems to be hampered by a lack of reproducibility. Variations in conductivity readings makes it very difficult, if not impossible, to define generally applicable diagnostic thresholds for discriminating among carious involvement. It may be hypothesized, that this problem is related to the very nature of the 2-point measurement configuration underlying all of the experimental methods, described in the literature as today. This study instead, aims to develop and evaluate a new method of spectroscopic Electrical Impedance Tomography (EIT), capable of reconstructing cross-sectional maps of the coronal tooth structure, depicting site-specific electrical impedance spectra. The resulting method promises to improve upon existing electrical caries detection methods, both in terms of its improved cross-sectional sampling strategy, dispensing with the need to rely on visual surface indications to determine appropriate measurement sites, and its immunity with regard to the natural variability in electrical conductance between individual teeth. Tomographic representation of the measurement results will allow the diagnostic interpretation to proceed on the basis of relative changes in tissue impedance among different spatial locations, instead of being dependent on a single quantitative interpretation. The suggested spectroscopic extension of the established method of EIT , together with the particularly difficult measurement conditions created by the insulating properties of the outer enamel layer, will pose some very specific problems as to the mathematical demands on the reconstruction algorithms. Therefore, the study and evaluation of non-linear reconstruction algorithms for the inverse conductivity problem of EIT, based on concepts drawn from numerical algebra, will constitute the major contribution in this research proposal. Despite the fact that evaluation of the algorithmic developments will be limited to numerical simulation and in vitro experiments, design of the newly developed method will be such, that its essential properties can be expected to transfer straightforwardly to the clinical setting. To gain a more fundamental understanding of the electrical conduction phenomena taking place in dental structures, a first, morphological accurate, numerical simulation model of a whole tooth will be built. The detailed morphological input data for this model will be derived from micro-Computed Tomography (micro-CT) scans, acquired with a microfocal desktop-sized measurement system. Cross-sectional images acquired in this way, will be processed by an automatic segmentation procedure, before being converted into a 3D Finite Element (FE) mesh. Such a mesh description allows to build up the actual volumetric simulation model.

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van de Wouwer Gert

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

This project studies new methods for texture feature extraction from digital images. We will emphasize the robustness of the features under influence of e.g. noise and illumination. In a first phase, these features will be used on publicly available image databases. In a second phase we will employ these techniques for the analysis of weak tissue tumors in NMR images.

Researcher(s)

• Promoter: Van de Wouwer Gert
• Co-promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The goal of the project is to develop methods for classification en detection based on hyperspectral images. Unmixing of the spectra into there constituent pure spectra is necessary to make the classification. Therefore, unmixing technique are developed to make this possible. The classification and unmixing techniques developed will be applied to the problem of vegetation.

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Co-promoter: Scheunders Paul
• Fellow: Tisson Greg

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Sijbers Jan
• Fellow: Leemans Alexander

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Fellow: Sijbers Jan

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk
• Fellow: D'haes Wim

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

The goal of this project is the development of a robust method for characterization of 3D binary objects, applied to rough diamonds. In particular, we will focus on the development of an invariant multiscale representation of a object surfaces (manifolds). It is expected that, at least scientifically, a possible solution will be provided for the 'blood diamonds' problem.

Researcher(s)

• Promoter: Sijbers Jan
• Co-promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Van Dyck Dirk

Research team(s)

• Vision lab

Project type(s)

• Research Project

Abstract

Researcher(s)

• Promoter: Scheunders Paul

Research team(s)

• Vision lab

Project type(s)

• Research Project