Deep Learning-based Automated Aortic Area and Distensibility Assessment: the Multi-Ethnic Study of Atherosclerosis (MESA).

Abstract

This study details application of deep learning for automatic segmentation of the ascending and descending aorta from 2D phase-contrast cine magnetic resonance imaging for automatic aortic analysis on the large MESA cohort with assessment on an external cohort of thoracic aortic aneurysm (TAA) patients. This study includes images and corresponding analysis of the ascending and descending aorta at the pulmonary artery bifurcation from the MESA study. Train, validation, and internal test sets consisted of 1123 studies (24,282 images), 374 studies (8067 images), and 375 studies (8069 images), respectively. The external test set of TAAs consisted of 37 studies (3224 images). CNN performance was evaluated utilizing a dice coefficient and concordance correlation coefficients (CCC) of geometric parameters. Dice coefficients were as high as 97.55% (CI: 97.47-97.62%) and 93.56% (CI: 84.63-96.68%) on the internal and external test of TAAs, respectively. CCC for maximum and minimum and ascending aortic area were 0.969 and 0.950, respectively, on the internal test set and 0.997 and 0.995, respectively, for the external test. The absolute differences between manual and deep learning segmentations for ascending and descending aortic distensibility were 0.0194 × 10-4 ± 9.67 × 10-4 and 0.002 ± 0.001mmHg-1, respectively, on the internal test set and 0.44 × 10-4 ± 20.4 × 10-4 and 0.002 ± 0.001mmHg-1, respectively, on the external test set. We successfully developed a U-Net-based aortic segmentation and analysis algorithm in both MESA and in external cases of TAA.