Multi-Task Deep Convolutional Neural Network for the Segmentation of Type B Aortic Dissection

Type B aortic dissection (TBAD) is a rare but life threatening disease. Segmentation of the entire aorta and truefalse lumen is crucial for the planning and follow-up of endovascular repair of TBAD. Manual segmentation in a slice-wise manner is time-consuming and requires expert experience. Current computer-aided methods have several limitations like focusing only on a specific part oftheaorta atatimeorrequiringhumaninteraction. Mostimportantly, these methods can not segment the entire aorta and detect true-false lumen at the same time. We report in this study a fully automatic approach based on multi-task deep convolutional neural network that segments the entire aorta and true-false lumen fromCTA images in a unified framework. Fortraining,webuiltadatabasecontaining254CTA images from both pre-operative and post-operative TBAD patients. These images are from multiple manufacturers. Slice-wise manual segmentation of the entire aorta and the true-false lumen for each 3-D CTA image is also provided. Our method is evaluated on 16 CTA data (11 preoperative and 5 postoperative) whose ground truth segmentation is provided by experienced vascular surgeons.Resultsshow that our method can segment type B aortic dissection with robustness and accuracy. Furthermore,our method can be easily extended to the segmentation of the entire aorta without dissection.