https://doi.org/10.1007/s00392-025-02625-4
Mani Arsalan (Frankfurt am Main)1, T. Duske (Mainz)2, H. Schneider (Frankfurt am Main)1, A. R. Tamm (Mainz)3, P. C. Seppelt (Frankfurt am Main)4, M. Geyer (Mainz)5, K. D. Piayda (Gießen)6, R. S. von Bardeleben (Mainz)2, D. Leistner (Frankfurt am Main)4, M. Hell (Mainz)5, T. Walther (Frankfurt am Main)1, F. Kreidel (Kiel)7
1Universitätsklinikum Frankfurt
Klinik für Thorax-, Herz- und Thorakale Gefäßchirurgie
Frankfurt am Main, Deutschland; 2Universitätsmedizin der Johannes Gutenberg-Universität Mainz
Zentrum für Kardiologie im Herz- und Gefäßzentrum
Mainz, Deutschland; 3Gesundheitscentrum Mainz Mitte
Interventionelle Kardiologie Sportmedizin Sportkardiologie
Mainz, Deutschland; 4Universitätsklinikum Frankfurt
Med. Klinik III - Kardiologie, Angiologie
Frankfurt am Main, Deutschland; 5Universitätsmedizin der Johannes Gutenberg-Universität Mainz
Kardiologie 1, Zentrum für Kardiologie
Mainz, Deutschland; 6Universitätsklinikum Gießen und Marburg GmbH
Medizinische Klinik I - Kardiologie und Angiologie
Gießen, Deutschland; 7Universitätsklinikum Schleswig-Holstein
Innere Medizin III mit den Schwerpunkten Kardiologie, Angiologie und internistische Intensivmedizin
Kiel, Deutschland
Background
Recently, a fully automated artificial intelligence (AI)-based computed tomography (CT) analysis software became available to simplify pre-procedural planning for transcatheter aortic valve implantation (TAVI). Precise measurements are of upmost importance for selecting the optimal valve prosthesis and thus achieving excellent clinical results. The aim of this study was to compare the semi-automated 3mensio (3M) software to the fully-automated Laralab platform (AI).
Methods
Patients with symptomatic severe aortic stenosis undergoing TAVI were retrospectively enrolled in two heart centres. The pre-procedural data set was analysed conventionally with the semi-automated, and the AI-based software package. We retrospectively simulated prothesis size choice based on the AI measurement and compared findings to the conventional valve sizing.
Results
247 patients were included in the analysis. The average annulus diameter was 24.5±2.3 mm (3M) vs. 24.4±2.4 mm (AI), the Mean Absolute Error (MAE) was 0.6 mm, the Mean Absolute Percentage Error (MAPE) was 2.6%, respectively. The mean annulus perimeter was 77.7±7.0 mm (3M) vs. 77.0±7.4 mm (AI), MAE: 2.0 mm, MAPE: 2.6%. The mean annulus area corresponded to 470.4±85.4 mm2 (3M) vs. 467.1±89.2 mm2 (AI), MAE: 21.2 mm2, MAPE: 4.6%. The intraclass correlation coefficients (ICCs) of all above mentioned parameters was >0.95, thus showing an excellent correlation between the two methods.
Prosthesis selection based on AI measurements alone would have resulted in a different valve size than the one actually implanted in 21% of patients. However, a simulation of prosthesis selection based on 3M measurements alone showed similar results with a different valve size in 14% of cases. (Figure 1)
Conclusions
In this retrospective analysis, an AI-based analysis of pre-procedural CT data sets showed excellent correlation with conventional measurement of the aortic annulus. However, like the semi-automated analysis, experienced users are still required to take all anatomical conditions into account in order to ensure optimal valve and valve size selection.