Longitudinal Segmentation of Multiple Sclerosis Lesions Using nnU-Net Architecture

Multiple sclerosis is a chronic autoimmune disease that affects the central nervous systems. The detection of new lesions through conventional magnetic resonance imaging is particularly important in the management of people with multiple sclerosis. The advancements in machine learning technology in recent years have significantly transformed the analysis of medical images for multiple sclerosis, particularly in identifying and segmenting lesions., improving the accuracy and efficiency of this process. In this context, the objective of this work is to develop a system for the detection of new lesions in people with multiple sclerosis using consecutive magnetic resonance scans. The proposed system uses only pre-processed FLAIR images to train a nnU-Net, a type of deep learning architecture that has been proven very successful for image segmentation tasks. The resulting model is able to generate masks that highlight changes between baseline and follow-up images, identifying new lesions that may have appeared in the meantime. The model achieved an average Dice score coefficient of 0.58 on the evaluation set. Overall, this work demonstrates the potential of machine learning tools for improving the detection and monitoring of multiple sclerosis lesions in clinical practice, particularly in the context of longitudinal studies.