Automated analysis of cardiac computed tomography (CCT) studies may help in personalized management and outcome prediction of patients undergoing Transcatheter aortic valve replacement (TAVR). This study aims to develop an object detection model and propose statistical algorithms to localize aortic root aligned with its tilt angle.

All consecutive patients who underwent CCT for TAVR procedure, from January to July 2023 at our center with no prior aortic valve prosthesis or permanent pacemaker were retrospectively included. The baseline annotations were performed by two experts. For object detection, MedYOLO, an adaptation of YOLO models for 3D volumetric data, was fine-tuned on the training set and its performance was evaluated by recall, precision, F1 and average precision at an IoU overlap of 50% (mAP50) and mAP50-95 on an unseen test set. For tilt alignment, intensity thresholding, connected component and principal component analysis were proposed and evaluated by Bland-Altman comparison.

Out of 179 CCTs, 100 CCTs were assigned to the test set, and the remaining to the training and validation using a 4:1 split. The model detected the aortic root with recall, precision and F1 score of 99.0%, respectively, mAP50 of 99.5% and mAP50-95 of 60.4%. The tilt prediction algorithm had a mean error of 7.9 [(-5.3)–21.1] degrees compared to the actual measurements and 3.3 [(-6.7)-13.4] between observers. 

This study demonstrates the robust performance of a fully automated pipeline for detection and analysis of key features in pre-TAVR CCTs. Further prospective studies are required for clinical developments.