Meeting Banner
Abstract #0531

Geometric and clinical evaluation of deep learning-based segmentation of individual lower limb muscles from patients with neuropathies

Marc-Adrien Hostin1,2, Augustin C. Ogier2, Constance P. Michel1, Yann Le Fur1, Maxime Guye1,3, Shahram Attarian4, Marc-Emmanuel Bellemare2, and David Bendahan1
1Aix Marseille Univ, CNRS, CRMBM, UMR 7339, Marseille, France, 2Aix Marseille Univ, Université de Toulon, CNRS, LIS, Marseille, France, 3APHM, Hopital Universitaire Timone, CEMEREM, Marseille, France, 4Centre de référence des maladies neuromusculaires et de la SLA, Marseille, France


Quantification of Fat Fraction (FF) in individual lower limb muscles of patients with neuromuscular disorders relies on segmentation. Few studies have indicated that Fully Convolutional Networks (FCNs) can provide reliable automatic segmentations to replace manual tasks. However, their sensitivity to fat infiltration has never been accurately assessed. Four FCN were benchmarked for the segmentation of 114 thigh and 108 calf images (1788 muscles) with FF up to 60%. HRNet was the only network that didn't show any segmentation failures. DSC obtained was comparable to other networks. The FF values calculated from the automatic (FFa) and manual (FFm) segmentations were consistent.

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

Join Here