精度TAVR定量化-AI加速TAVR計画により二尖弁大動脈弁狭窄症の評価時間が80%短縮

Bicuspid aortic valve (BAV) stenosis complicates transcatheter aortic valve replacement (TAVR) planning, with no validated automated measurement algorithm available. We developed Cardioverse, the first fully automated deep learning algorithm for BAV anatomical assessment in TAVR planning.

We conducted a large-scale, multicenter retrospective study encompassing 1,147 consecutive patients with BAV undergoing TAVR across 16 high-volume Chinese centers (March 2019-February 2023). Cardioverse was trained on this cohort and evaluated in an internal (n = 437) and external (n = 110) validation cohorts. Our novel Cardioverse algorithm demonstrated exceptional segmentation performance across all anatomical targets with Dice similarity coefficients >0.97 for coronary cusps and ostia. Critically, the algorithm achieved unprecedented workflow efficiency gains: 80% reduction in assessment time [241.0 IQR (181.0, 297.0) vs. 1251.0 IQR (872.0, 1408.0) seconds], 85% reduction in user interactions [57.0 IQR (45.0, 78.0) vs. 382.5 IQR (285.5, 475.0) clicks], and 87% reduction in manual effort [7.5 IQR (4.8, 9.3) vs. 57.2 IQR (43.5, 68.4) meters mouse movement] compared to expert observers (P < 0.001). Importantly, accuracy was maintained across all BAV phenotypes with correlation coefficients >0.91 for all critical measurements, including annular dimensions, calcification quantification, and aortic root morphology assessment.

Cardioverse transforms pre-TAVR assessment for BAV patients, offering a validated solution combining accuracy and efficiency. It reduces assessment time from over 20 min to < 5 min, addressing the need for standardized, rapid, and reliable BAV evaluation. Its robust performance across diverse BAV phenotypes makes it a crucial tool for enhancing TAVR planning consistency.

ClinicalTrials.gov Protocol Registration System (NCT05044338).