Transformer-Driven segmentation of New Lesions in Multiple Sclerosis scans

Multiple sclerosis (MS) is a chronic neurological disorder characterized by the development of lesions in the central nervous system, typically occurring in the brain whose diagnosis and efficacy of treatment monitoring are vital. Timely detection of new lesions is crucial to analyse disease progression and evaluate treatment efficacy. In this study, we use a novel deep learning model; ChangeFormer, for automated detection of the new and enlarging lesions in brain magnetic resonance imaging (MRI) data of MS patients. ChangeFormer utilizes a transformer-based Siamese architecture, leveraging the concept of spacial change detection we applied the model to identify the change in MRI scans, appearing as MS lesions. We evaluate the performance of the model on a publicly available MS dataset and compare it with state-of-the-art methods. Our experimental results demonstrate the effectiveness of ChangeFormer in accurately detecting new and enlarging lesions, showcasing its potential as a valuable tool in clinical practice for monitoring disease activity in MS patients.