Multiple sclerosis (MS) is a disease in which the myelin sheets covering the nerve cells in the central nervous system (i.e., brain and spinal cord) are damaged. Such demyelination disrupts the communications along the nerves, leading to a number of physical and mental impairments. MS is the most common autoimmune disorder related to the central nervous system, but there is unfortunately no cure for it, and can lead to permanent disability and death. During the last 10-20 years, computerized image analysis of human brain magnetic resonance imaging (MRI) scans has been a major topic of research within medical imaging. This research has led to many software packages (e.g., FSL, FreeSurfer, SPM), which are extensively used in neuroimaging studies. Moreover, there is a wide literature on the analysis of brain MRI of MS patients, mostly focused on the segmentation and quantification of white matter lesions. However, image analysis of the spinal cord has received considerably less attention, despite its clinical importance in MS. The only publicly available package is SpinalCordToolbox (SCT), which is able to extract the spinal cord from MRI, parcellate it, and place it in a reference coordinate system. However, SCT has two limitations that prevent its application in longitudinal studies of MS: 1. It is not robust against pathology; and 2. It cannot handle longitudinal data. It is the goal of this PhD project to overcome these limitations and create tools that can be applied to longitudinal studies of MS, enabling us to better characterize the disease.
Aims and Objectives:
The project has two main aims: Aim 1: creating an image analysis pipeline that is robust against MS pathology. One possibility will be to segment the MS lesions, inpaint them, and then use standard tools (e.g., SCT). We will also develop methods for spinal cord segmentation, centerline extraction, and registration that can handle the changes in image appearance caused by MS – producing lesion segmentations as a by-product. Aim 2: creating a longitudinal version of the pipeline in Aim 1. By exploiting the knowledge that a set of scans are from the same subject, we can obtain more robust results than when naively analyzing each time point independently. At the core of the method will be a subject-specific atlas, connecting the time points with one another and with a global atlas, to produce robust estimates of spinal cord atrophy and MS lesion changes in different populations (e.g., treatment vs. no treatment).
The University College London EPSRC Centre for Doctoral Training in Medical Imaging
