Abstract #0041

Breaking the clinico-radiological paradox in multiple sclerosis using machine learning

Arnaud Attyé^1,2, Stenzel Cackowski³, Alan Tucholka⁴, Pauline Roca⁴, Pascal Rubini⁴, Sebastien Verclytte⁵, Lucie Colas⁵, Juliette Ding⁵, Jean-François Budzik⁵, Felix Renard⁶, Emmanuel L Barbier³, Romain Casey^7,8,9,10, Sandra Vukusic^7,8, and François Cotton^7,11

¹Grenoble alpes university, Grenoble, France, ²Sydney Imaging Lab, Sydney university, Sydney, Australia, ³Univ. Grenoble Alpes, Inserm, U1216, Grenoble Institute Neurosciences, Grenoble, France, ⁴Pixyl Medical, Grenoble, France, ⁵Lille Catholic University, Lille, France, ⁶Laboratoire d'informatique de Grenoble, Grenoble, France, ⁷Claude Bernard Lyon 1 University, Lyon, France, ⁸Lyon University Hospital, Lyon, France, ⁹Observatoire Français de la Sclérose en Plaques, INSERM 1028 et CNRS UMR 5292, Lyon, France, ¹⁰EUGENE DEVIC EDMUS Foundation against multiple sclerosis, Lyon, France, ¹¹CREATIS, CNRS UMR 5220 - INSERM U1206, Lyon, France

MRI is central to the study of white matter lesions in multiple sclerosis (MS). To date, the distribution of MS lesions, as evaluated on FLAIR imaging, has not been linked to patients’ disability prediction. Based on an international data challenge with 1500 MS patients and ground truth 2-year Expanded Disability Status Scale (EDSS), we have proposed an adaptive machine learning framework to predict the clinical disability. Here, we report the encouraging finding that our algorithm predicts the 2-year EDSS score with an accuracy estimated to 81%, only based on a single initial FLAIR sequence, added to sex and gender information.

This abstract and the presentation materials are available to members only; a login is required.

Join Here