Objective:To explore the construction of an evaluation model for aortic root anatomy and calcium burden in patients with bicuspid aortic valve (BAV) stenosis before transcatheter aortic valve replacement (TAVR) based on deep learning (DL) algorithms.Methods:A retrospective collection of 362 BAV stenosis patients who underwent TAVR from September 2023 to May 2024 was performed. All patients underwent cardiac CT angiography. The patients were divided into training group ( n=104), internal validation group ( n=206), and external validation group ( n=52). A DL model was trained on the training dataset to assess aortic root anatomy and calcification burden. The evaluation included the segmentation accuracy of the algorithm, the measurement performance of key anatomical structures (i.e., valve leaflets and type-1 and type-2 fusion raphe), and calcification burden, as well as the measurement efficiency. Overall segmentation performance was assessed using the average Dice coefficient (ADC). The fine-scale segmentation quality was validated by the 95th-percentile Hausdorff distance (HD-95) and the average symmetric surface distance (ASSD). The consistency of the measurement results was assessed using the Pearson correlation coefficient and the intraclass correlation coefficient ( ICC) with a two-way mixed model for absolute agreement. In addition, the total time and total mouse movement distance required for manual assessment versus the DL model on the validation datasets were recorded and compared. Results:The algorithm demonstrated excellent segmentation performance on aortic root anatomical targets, achieving outstanding consistency within both internal and external validation datasets (0.955

Objective:To explore the construction of an evaluation model for aortic root anatomy and calcium burden in patients with bicuspid aortic valve (BAV) stenosis before transcatheter aortic valve replacement (TAVR) based on deep learning (DL) algorithms.Methods:A retrospective collection of 362 BAV stenosis patients who underwent TAVR from September 2023 to May 2024 was performed. All patients underwent cardiac CT angiography. The patients were divided into training group ( n=104), internal validation group ( n=206), and external validation group ( n=52). A DL model was trained on the training dataset to assess aortic root anatomy and calcification burden. The evaluation included the segmentation accuracy of the algorithm, the measurement performance of key anatomical structures (i.e., valve leaflets and type-1 and type-2 fusion raphe), and calcification burden, as well as the measurement efficiency. Overall segmentation performance was assessed using the average Dice coefficient (ADC). The fine-scale segmentation quality was validated by the 95th-percentile Hausdorff distance (HD-95) and the average symmetric surface distance (ASSD). The consistency of the measurement results was assessed using the Pearson correlation coefficient and the intraclass correlation coefficient ( ICC) with a two-way mixed model for absolute agreement. In addition, the total time and total mouse movement distance required for manual assessment versus the DL model on the validation datasets were recorded and compared. Results:The algorithm demonstrated excellent segmentation performance on aortic root anatomical targets, achieving outstanding consistency within both internal and external validation datasets (0.955