Research team
Expertise
I perform research on motion correction methods and image reconstruction for positron emission tomography (PET). Specifically, I develop motion tracking and reconstruction methods for brain PET scans experiments of awake/freely moving animals. These methods have also been adapted for motion tracking in human brain PET scans. In addition, I develop PET reconstruction approaches to improve image resolution. The raw PET tomographic data is processed and motion tracking algorithms are implemented to measure the motion of the subject during the PET study. Once the motion tracking data has been calculated, it is used in the PET motion correction reconstruction methods to obtain tomographic images without motion artifacts. I have performed brain PET scan experiments of awake rats, in which head motion tracking methods were developed and tested: Marker based optical motion tracking, markerless optical motion tracking, and radioactive point sources motion tracking were considered. The point source tracking method was also implemented in brain PET scans of freely running mice. The point source tracking method has been adapted to a dedicated human brain PET scanner to track the head motion of patients to reduce motion artefacts that occur during longer PET scan sessions.
Machine learning based motion tracking for awake brain PET scans in a high-resolution scanner.
Abstract
Preclinical positron emission tomography (PET) is a unique in vivo imaging technique that allows to study the distribution of radiolabeled biomolecules in the whole body of small animals. To avoid motion during the PET scan, animals are anesthetized, but for brain studies the effect of anesthesia can represent a confounding factor. Anesthesia has been shown to change the binding of radiotracers targeting neurotransmitter receptors, change glucose metabolism and alter cerebral blood flow. To avoid these confounding effects, methods to allow scanning of awake, unanesthetized, animals are being developed. Awake scanning can be done by tracking the head motion of the animal during the scan to then perform motion correction reconstruction. The practicality of the tracking method is crucial to allow routine implementation of awake PET scans. In this project we will develop a head motion tracking method for awake PET scans which relies only on the PET activity of the scan itself (i.e. data-driven), therefore eliminating the need for additional tracking hardware or extra animal preparation for the scans. We expect that the practicality of this approach will allow to perform awake PET scans routinely. The data-driven motion tracking algorithm consists of dividing the PET data in short time frames of 32 ms, which have null or little motion, followed by calculation of the animal head position in all these frames. To determine the head position, the animal body is segmented, but due to the high noise in the short time frames, the segmentation can be inaccurate. In this project, we will implement and optimize a residual recurrent U-net neural network to increase the accuracy of the animal body segmentation. For training, we will use aligned PET-CT images of anesthetized animals, where the label is the short frame PET reconstruction and the ground-truth segmented body can be obtained from the CT image. Then, in the awake scans, where only the PET activity is available, the segmented body will be inferred with the U-net. This network has been used in other medical image segmentation tasks with great success. Thus, using machine learning we expect to greatly improve the accuracy of the data-driven motion tracking, and therefore to obtain higher quality motion corrected images in awake PET scans. Since these methods will be developed in a new generation, high resolution, preclinical PET scanner recently acquired in our lab, the first objective of this project is to adapt the reconstruction algorithms developed in our old scanner to the new scanner, to then develop the data-driven motion tracking as described above. Implementing a practical method to perform awake brain PET scans in a new generation scanner will help to adopt this technology routinely, not only in our lab, but by the larger preclinical research community.Researcher(s)
- Promoter: Miranda Menchaca Alan
Research team(s)
Project type(s)
- Research Project
Data-driven brain and heart PET imaging of awake rodents.
Abstract
Preclinical brain positron emission tomography (PET) is performed using anaesthetics to immobilize the animal. However, it has been shown that anaesthetics can influence brain function and the uptake of several PET tracers. To circumvent the use of anaesthesia, methods that track the motion of the animal head during the PET scan have been developed, which aid subsequent motion correction of the brain PET data. These methods rely on optical tracking cameras or on markers attached on the animal head to track the head motion. Therefore, the tracking procedure requires additional setup procedures in addition to the PET scan itself. To improve practicality and to reduce the additional setup to perform scans of awake rodents the minimum, we will validate and optimise a rodent head data-driven tracking technique in this project. This method requires no additional setup in addition to the PET scan, since the motion tracking is performed using the acquired PET data. Additionally, a torso motion tracking will be validated and optimised to perform heart motion correction reconstruction. The heart image can then be used to obtain the image derived input function to perform improved quantification with kinetic modelling. Rat and mice scans will be performed with different PET tracers to obtain data with different brain and body distributions, as well as different noise characteristics, to optimise the tracking algorithms. The methods developed here will serve to improve practicality of awake PET rodent scans, therefore facilitating adoption of the technique by the wider PET preclinical community. By circumventing the use of anaesthetics and their confounding effects on the animal physiology, translation of preclinical PET results to the clinic will be improved.Researcher(s)
- Promoter: Miranda Menchaca Alan
Research team(s)
Project type(s)
- Research Project