Multiple System Atrophy Classification via 3D Convolutional Neural Network and Simulated Brain MRI Parametric Maps
Giulia Maria Mattia1, Edouard Villain1,2, Olivier Rascol3, Wassilios G. Meissner4,5,6, Xavier Franceries7, and Patrice Péran1
1ToNIC, Toulouse NeuroImaging Center, Université de Toulouse, Inserm, UPS, Toulouse, France, 2LAAS CNRS, Université de Toulouse, CNRS, INSA, UPS, Toulouse, France, 3French Reference Center for MSA, Department of Neurosciences and Clinical Pharmacology Clinical Investigation Center CIC1436, NS-Park/FCRIN Network, ToNIC, Toulouse NeuroImaging Center University Hospital of Toulouse, Inserm, Université de Toulouse 3, Toulouse, France, 4French Reference Center for MSA, Department of Neurology for Neurodegenerative Diseases, University Hospital Bordeaux, Bordeaux, France, 5Univ. Bordeaux, CNRS, IMN, UMR 5293, F-33000 Bordeaux, France, 6Dept. Medicine, University of Otago, Christchurch and New Zealand Brain Research Institute, Christchurch, New Zealand, 7CRCT, Centre de Recherche en Cancerologie de Toulouse, Inserm, Toulouse, France
As a rare neurodegenerative disorder, multiple system atrophy (MSA) can be challenging to analyse using a deep learning approach given the limited sample size. A method is submitted to produce cluster-based simulated brain MR parametric maps from healthy controls, using regional intensity distribution belonging to a set of MSA patients. This enabled to train a 3D CNN only with the simulated set. Testing on the MSA data set, the accuracy obtained was comparable to the state-of-the-art. This approach allows to deal with small samples of data in deep learning, while exploiting a-priori knowledge of the disease.
This abstract and the presentation materials are available to members only;
a login is required.