Deep learning (DL) is an effective way for performing automatic multi-channel (or contrast) semantic segmentation. Here we investigated the accuracy of tissue segmentation as a function of the number and combinations of contrasts to the input of a fully convolutional neural network. The multi-contrast images included FLAIR, pre-contrast T1-, T2-, and proton density-weighted images, acquired on a large cohort of multiple sclerosis patients. Our results show that the number of input channels affects the segmentation accuracy in a tissue-dependent manner and that FLAIR is the major determinant of segmentation accuracy.
This abstract and the presentation materials are available to members only; a login is required.