Improving contour detection in MS white matter lesions using the computer vision methods of morphological snakes

Details

Supervisors

Macq, Benoît

Faculty

Ecole polytechnique de Louvain

Degree label

Master [60] en sciences informatiques

Abstract

Multiple Sclerosis (MS) is a chronic disease that affects the central nervous system, leading to the formation of lesions, which are areas of damaged tissue in the brain and spinal cord. Accurate knowledge of lesion size and count is important for diagnosis and monitoring the progression of MS. This research aims to improve the segmentation of MS lesions and detect confluent lesions, which are larger areas where multiple lesions have merged and are more challenging to identify. The study enhances existing models like 3D Unet + CC, 3D Unet + ACLS, and ConfLUNet using post-processing techniques known as morphological snakes. Despite initial expectations, the application of these techniques did not significantly improve segmentation accuracy, particularly for smaller lesions. Challenges were also noted in the accurate detection and segmentation of confluent lesions. The study highlights the difficulties in achieving consistent segmentation results across different lesion sizes when using morphological snakes and suggests that future research should explore more refined methods and test them on additional models to improve diagnostic accuracy.