This repository contains : Complete research code for analysis and post-hoc experiments related to multi-centric cortical lesion segmentation models in Multiple Sclerosis MRI The complete research code related to the ArXiV: arXiv:2507.12092 Research Overview:This work presents a comprehensive benchmarking study of deep learning approaches for cortical lesion detection in multiple sclerosis, with focus on multi-site generalization and model explainability. Cortical lesions are valuable biomarkers offering high diagnostic specificity, but their clinical integration remains limited due to imaging challenges and lack of standardized automated methods. Key Features: Multi-site validation across four major medical centers: University Hospital Basel (INsIDER dataset, 3T), Lausanne University Hospital - CHUV (Advanced dataset, 3T), National Institutes of Health (NIH, 3T and 7T), and University of Louvain (UCLouvain dataset, 3T) nnU-Net-based¹ deep learning pipeline optimized for cortical lesion segmentation Cross-domain validation and performance explanation framework Stratified evaluation considering lesion characteristics and site diversity Complete preprocessing pipeline with SynthStrip² integration Latent space analysis for model interpretability Model Specifications: Architecture: nnU-Net 3D full resolution Input: T1-weighted MRI (MP2RAGE/MPRAGE sequences) Output: Binary cortical lesion probability maps Training: Multi-site MS cohort with stratified cross-validation Ready-to-Use Docker Container:Pre-trained model available at: dockerhub Citation: This work is described in: "Benchmarking and Explaining Deep Learning Cortical Lesion MRI Segmentation in Multiple Sclerosis" (arXiv:2507.12092, 2025). When using this code or the Docker, please cite both this software repository (https://doi.org/10.5281/zenodo.15911797) and the associated research paper "Benchmarking and Explaining Deep Learning Cortical Lesion MRI Segmentation in Multiple Sclerosis", N. Molchanova, A. Cagol, M. Ocampo-Pineda, P-J Lu, M. Weigel, X. Chen, E. Beck, C. Tsagkas, D. Reich, C. Vanden Bulcke, A. Stolting, S. Borrelli, P. Maggi, A. Depeursinge, C. Granziera, H. Mueller, P. M. Gordaliza, M. Bach Cuadra (arXiv:2507.12092, 2025). References: [1] Isensee, Fabian, Paul F. Jaeger, Simon A.A. Kohl, Jens Petersen, and Klaus H. Maier-Hein. “nnU-Net: A Self-Configuring Method for Deep Learning-Based Biomedical Image Segmentation.” Nature Methods 18, no. 2 (2021): 203–11. https://doi.org/10.1038/s41592-020-01008-z. [2] Hoopes, Andrew, Jocelyn S. Mora, Adrian V. Dalca, Bruce Fischl, and Malte Hoffmann. “SynthStrip: Skull-Stripping for Any Brain Image.” NeuroImage 260 (October 2022): 119474. https://doi.org/10.1016/j.neuroimage.2022.119474.