Prof. dr. Pieter Van Dyck is head clinic radiology at the Antwerp University Hospital (UZA) specialized in musculoskeletal (MSK) imaging. He has large clinical expertise in conventional radiography (XR), ultrasound (US), computer tomography (CT) and magnetic resonance (MRI) imaging of the MSK system. Clinical research in the domain of new and advanced MRI techniques for MSK applications in collaboration with the Imec-Vision Lab (UAntwerpen). Emphasis on the development, clinical validation and implementation of super-resolution reconstruction (SRR) techniques for anatomical and quantitative MRI of the knee.
Development of algorithms to predict outcome after mild traumatic brain injury using magnetic resonance imaging.
AbstractTraumatic Brain Injury (TBI) is a sudden damage in the brain caused by an external force to the head. There are 50 million TBI cases worldwide every year, with over 2.5 million people in Europe. Mild TBI (mTBI) cases account for over 85% of the head injuries. In up to 40% of the cases, recovery from mTBI may be incomplete, with patients having persistent motor and psychological impairment for months to years after injury. Magnetic resonance imaging (MRI) is a helpful tool for documenting the extent of brain damage and is routinely applied in clinical practice. However, patients with mTBI often do not show abnormalities on conventional MRI, despite changes in behavior or cognitive deficits. This observation indicates that some microstructural changes in the brain cannot be detected by conventional MRI, but may be important determinants of a patient's clinical outcome. A number of studies using more advanced diffusion MRI (dMRI) have reported changes in diffusion parameters associated with meaningful clinical measures, such as cognitive and functional impairment in mTBI. Currently, radiologists and neurologists cannot predict the outcome of mTBI patients based on conventional MR images. However, recent developments in machine learning methods have shown promise for improving outcome prediction by combining multiple clinical parameters. We have previously explored Support Vector Machine learning algorithms for outcome prediction of mTBI patients based on voxel-wise analysis of FLAIR, SWI, FA and MD images. In the next step, we aim to explore Deep Learning algorithms for outcome prediction. Deep learning is facilitated by neural networks that mimic the neurons in the human brain, which autonomous learns from data. Specifically, we will implement a convolutional neural networks (CNN) based methods for outcome prediction in mTBI, using conventional and diffusion MRI from the multi-site CENTER-TBI project. Good ve\rsus incomplete recovery will be dichotomised using the Extended Glasgow Coma Scale (GOSE) score, evaluated at 6 months after injury. Ultimately, improved prediction of mTBI outcome would help to stratify patients and better organize the management of mTBI patients with poor outcome.
- Promotor: Van Dyck Pieter