Automatic Rodent Brain MRI Lesion Segmentation with Fully Convolutional Networks

• DOI: 10.48550/arxiv.1908.08746
• Publication Date: 2019-01-01
• Authors: Juan Miguel Valverde; Artem Shatillo; Riccardo De Feo; Olli Gröhn; Alejandra Sierra; Jussi Tohka
• Keywords: Sørensen–Dice coefficient; Dice

Manual segmentation of rodent brain lesions from magnetic resonance images (MRIs) is an arduous, time-consuming and subjective task that highly important in pre-clinical research. Several automatic methods have been developed for different human MRI segmentation, but little research has targeted lesion segmentation. The existing tools performing rodents are constrained by strict assumptions about the data. Deep learning successfully used medical image However, there not any deep approach specifically designed tackling In this work, we propose a novel Fully Convolutional Network (FCN), RatLesNet, aforementioned task. Our dataset consists 131 T2-weighted rat scans 4 studies which ischemic stroke was induced transient middle cerebral artery occlusion. We compare our method with two other 3D FCNs originally anatomical (VoxResNet 3D-U-Net) 5-fold cross-validation on single study generalization test, where training done testing three remaining studies. labels generated were quantitatively qualitatively better than predictions compared methods. average Dice coefficient achieved experiment proposed 0.88, between 3.7% 38% higher architectures. presented architecture also outperformed at generalizing studies, achieving 0.79.