Abstract 3652: Computer-Aided Diagnosis of Congenital Aortic Disease using Aortic Shape and Motion
Early detection of aortic aneurysms is vital to reduce morbidity and mortality of patients with Marfan connective tissue disease. We report a computer-aided diagnosis (CAD) system for objective and early identification of aortic aneurysm from 4-dimensional (3D+time) aortic MR images. The developed method is fully automated and consists of 4D aortic segmentation, shape/motion description, and connective tissue disease determination. A 4D segmentation algorithm segments the luminal aortic surfaces in all 16 cardiac phases using a single optimal segmentation process after fusing 2D cine SSFP MR candy-cane and LVOT 4D images (1.5–2.0 mm3 voxel). Principal component analysis (PCA) applied to aortic surfaces produces compact indices of aortic shape and motion. A support vector machine classifier is used to assign the aortic MR data in two classes of normal and connective tissue disease subjects. The method’s performance was assessed in 16-phase aortic MR images from 36 subjects (20 normal, 16 patients) using a leave-one-out training/testing approach. The patients were clinically identified based on their familial history and did not show qualitative visual evidence of aortic disease. The performance was evaluated in terms of the overall classification correctness and expressed in percent. Two different CAD tasks were performed based on single end-diastolic 3D aortic shape (current clinical standard), and full 16-phase aortic shape/motion. The computer-determined aortic surface covered the entire thoracic aorta, yielding subvoxel segmentation accuracy with signed positioning errors of −0.10±2.05 mm. The two CAD tasks correctly discriminated patients from normal subjects in 86.1% for a 3D analysis and in 94.4% for 16-phase 4D analysis. The classification correctness of the 16-phase 4D analysis is significantly better than that of the single-phase 3D approach (p<0.05). Our novel CAD technique shows that aortic motion information, lacking in the 3D model but present in the 16-phase 4D model, is useful in the detection of Marfan connective tissue disease and the associated changes of aortic shape and motion. While the differences in aortic motion are difficult, if not impossible, to observe visually, they are reliably detected by the reported CAD system.