Maxence Wynen1, Francesco La Rosa2,3,4, Amina Sellimi5, Germán Barquero2,3,4, Gaetano Perrotta6, Valentina Lolli7, Vincent Van Pesch5, Cristina Granziera8,9, Tobias Kober10, Pascal Sati11,12, Benoît Macq13, Daniel S. Reich11, Martina Absinta11,14, Meritxell Bach Cuadra2,3,4, and Pietro Maggi5,15
1Ecole Polytechnique de Louvain, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, 2Signal Processing Laboratory (LTS5), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 3Medical Image Analysis Laboratory, Center for Biomedical Imaging (CIBM), University of Lausanne, Lausanne, Switzerland, 4Radiology Department, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, 5Department of Neurology, Cliniques Universitaires Saint-Luc, Université Catholique de Louvain, Brussels, Belgium, 6Department of Neurology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, 7Department of Radiology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, 8Neurologic Clinic and Policlinic, Departments of Medicine, Clinical Research and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 9Translational Imaging in Neurology (ThINk) Basel, Department of Medicine and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 10Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 11Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, United States, 12Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 13ICTEAM Institute, Université catholique de Louvain, Louvain-la-Neuve, Belgium, 14Department of Neurology, Johns Hopkins University, Baltimore, MD, United States, 15Department of Neurology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland
1Ecole Polytechnique de Louvain, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, 2Signal Processing Laboratory (LTS5), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 3Medical Image Analysis Laboratory, Center for Biomedical Imaging (CIBM), University of Lausanne, Lausanne, Switzerland, 4Radiology Department, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, 5Department of Neurology, Cliniques Universitaires Saint-Luc, Université Catholique de Louvain, Brussels, Belgium, 6Department of Neurology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, 7Department of Radiology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, 8Neurologic Clinic and Policlinic, Departments of Medicine, Clinical Research and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 9Translational Imaging in Neurology (ThINk) Basel, Department of Medicine and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 10Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 11Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, United States, 12Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 13ICTEAM Institute, Université catholique de Louvain, Louvain-la-Neuve, Belgium, 14Department of Neurology, Johns Hopkins University, Baltimore, MD, United States, 15Department of Neurology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland
RimNet, a deep-learning-based segmentation
algorithm for chronic active multiple sclerosis lesions, is effective for unseen
MRI data acquired on a scanner from a different vendor, and makes consistent
predictions on longitudinal data.
Figure 1: Representative 3D FLAIR* and Phase images from one
PMS patient showing the same PRL a) at baseline and b) at the 13 months post-DMT
follow-up.
Figure 3: Summary of the lesion-wise results for each patient with paramagnetic rim lesions (n=11 of 13 total patients studied, 85%). a) Number of rims detected in session 1 (baseline) and session 2 (follow-up) for each patient, overlaid on the total number of rims adjudicated in the consensus manual reading. b) Number of matching predictions between baseline and follow-up. Abbreviations: Consistencyp, probability-based consistency; Consistencyb, binary-based consistency.
