Multi-Class Ct Segmentation Of Abdominal Aortic Aneurysm Including Thrombotic Masses And Calcification

Early prediction of rupture risk for the abdominal aortic aneurysm significantly reduces mortality rates. An accurate automatic segmentation applied to CT images is of utmost importance to facilitate medical routines and guarantee low level of detection errors. We address this issue employing a 3D-based FCNN. Due to the biomechanical importance for the blood flow simulations, the main emphasis in the present work is on calcification areas, which are heavily under-represented compared to the aortic lumen domain and thrombotic masses. We collect a CT dataset of 30 patients annotated by 3 medical experts. The dice score (volumetric similarity) for the aortic lumen, thrombotic masses, and calcifications reaches so far the accuracy of about 96% (98%), 80% (90%) and 66% (86%), respectively. These results allow us to recover the lumen geometry together with accurate elastic properties of the aortic wall governed by thrombotic masses and calcification distributions for hemodynamic simulations.