Here, we trained and evaluated a fully convolutional neural network for 3D images to automatically segment brain lesions in MRI images of a cuprizone mouse model of multiple sclerosis. To improve performance, several pre-processing and data augmentation methods were tested and compared. The impact of lesion size on network performance was evaluated and we applied the trained segmentation model to images from non-lesion control mice to assess the capacity of the network to detect the presence or absence of lesions. We conclude that the trained network can 1) detect the presence of a lesion and 2) accurately segment the volume.
This abstract and the presentation materials are available to members only; a login is required.