Segmentation of Traumatic Brain Injuries with Convolutional Neural Networks

Objective: In this work we investigate the capabilities of Convolutional Neural Networks (CNN) for the segmentation of brain lesions. We present a fully automatic segmentation system, the design of which was driven by the most recent advances in the field of Deep Learning, adapted for the accurate segmentation of brain lesions in multi-modal MRI scans. We demonstrate that Deep Convolutional Neural Networks with carefully designed architecture can be successfully trained even on biomedical datasets of limited size and show very promising performance on the challenging task of TBI lesion segmentation. Our network was evaluated on two datasets comprising MRI scans of patients with severe TBI, where it surpassed the performance of previously developed methods.