A Fully Automated AI-Based Software for Structural Mitral Valve CT Analysis

Background: Transcatheter mitral valve replacement (TMVR) has emerged as treatment for severe mitral regurgitation (MR). Pre-procedural 4D-CT assessment is a crucial prerequisite to determine therapy eligibility of the patient’s anatomy and optimize valve selection. However, CT analysis and high screen-failure rates hinder broader applicability, resulting in extensive time and resource consumption. This study evaluates a novel fully automated AI-based software for TMVR CT screening, in comparison to conventional manual analysis.

Methods: Screening CTs for TMVR with a tether-based device at 4 high-volume centers were analyzed. Conventional manual core-laboratory screening (CTman) measurements were compared with a fully automated algorithm (CTai). Additionally, a facilitated grading system (green = feasible, yellow = potentially feasible, red = not feasible) was developed and tested.

Results: 181 CTs were analyzed, with excellent agreement between CTai and CTman measurements (Mitral annulus perimeter: 116.8 ± 17.4 [CTai] versus 114.2 ± 18.3 mm [CTman], intraclass correlation coefficient [ICC, 95%CI]: 0.937[0.911–0.954]; intercommissural distance: 39.0 ± 6.2 versus 38.9 ± 6.4 mm, ICC: 0.869 [0.838–0.894]; septolateral diameter: 29.7 ± 5.7 versus 29.1 ± 5.4 mm, ICC: 0.887 [0.859–0.909]; minimal endsystolic LVOT area: 341.6 ± 163.1 versus 326.0 ± 171.8 mm², ICC: 0.800 [0.725–0.855]). The CTai pre-screening process demonstrated high accuracy compared with core-laboratory recommendations (sensitivity: 91.6%, specificity: 94.2%, s. Graphic). 8% false negative and 5% false positive cases occurred mainly due to prior transcatheter edge-to-edge repair (TEER) and mitral annular calcification.

Conclusion: CTai resulted in reliable anatomic measurements compared with CTman. Additionally, the automatic and facilitated pre-screening tool showed adequate clinical accuracy. The integration of this fully automated, AI-based software into the pre-procedural screening process can lead to time and resource savings. Increasing data input and further algorithm refinements for more complex anatomies (e.g., post MV implantations or TEER devices) will further enhance accuracy.

NB: This abstract was presented in a similar form at a previous meeting.

Publikationsverlauf

Artikel online veröffentlicht:
11. Februar 2025