Evaluating the Performance of a Convolutional Neural Network Algorithm for Measuring Thoracic Aortic Diameters in a Heterogeneous Population

The purpose of this work was to assess the performance of a convolutional neural network (CNN) for automatic thoracic aortic measurements in a heterogeneous population. From June 2018 to May 2019, this study retrospectively analyzed 250 chest CT scans with or without contrast enhancement and electrocardiographic gating from a heterogeneous population with or without aortic pathologic findings. Aortic diameters at nine locations and maximum aortic diameter were measured manually and with an algorithm (Artificial Intelligence Rad Companion Chest CT prototype, Siemens Healthineers) by using a CNN. A total of 233 examinations performed with 15 scanners from three vendors in 233 patients (median age, 65 years [IQR, 54.72 years]; 144 men) were analyzed: 68 (29%) without pathologic findings, 72 (31%) with aneurysm, 51 (22%) with dissection, and 42 (18%) with repair. No evidence of a difference was observed in maximum aortic diameter between manual and automatic measurements (P =.48). Overall measurements displayed a bias of -1.5 mm and a coefficient of repeatability of 8.0 mm at BlandAltman analyses. Contrast enhancement, location, pathologic finding, and positioning inaccuracy negatively influenced reproducibility (P < .003). Sites with dissection or repair showed lower agreement than did sites without. The CNN performed well in measuring thoracic aortic diameters in a heterogeneous multivendor CT dataset. (C)RSNA, 2022