Abstract
Acute aortic syndrome (AAS) is a group of life threatening conditions of theaorta. We have developed an end-to-end automatic approach to detect AAS incomputed tomography (CT) images. Our approach consists of two steps. At first,we extract N cross sections along the segmented aorta centerline for each CTscan. These cross sections are stacked together to form a new volume which isthen classified using two different classifiers, a 3D convolutional neuralnetwork (3D CNN) and a multiple instance learning (MIL). We trained, validated,and compared two models on 2291 contrast CT volumes. We tested on a set asidecohort of 230 normal and 50 positive CT volumes. Our models detected AAS withan Area under Receiver Operating Characteristic curve (AUC) of 0.965 and 0.985using 3DCNN and MIL, respectively.