Impact of fully-automated AI based CT-analysis on pre-procedural TAVI planning.
Authors
Affiliations (8)
Affiliations (8)
- Department of Cardiac Surgery, University Hospital of the Goethe University, Theodor-Stern-Kai 7, 60590, Frankfurt/Main, Germany. [email protected].
- Department of Cardiology and Angiology, Medical Clinic I, University Hospital of the Justus Liebig University, Giessen, Germany. [email protected].
- Department of Cardiology, University Medical Centre Mainz, Mainz, Germany.
- Department of Cardiac Surgery, University Hospital of the Goethe University, Theodor-Stern-Kai 7, 60590, Frankfurt/Main, Germany.
- Department of Cardiology & Angiology, Medical Clinic III, University Hospital of the Goethe University, Frankfurt/Main, Germany.
- Department of Cardiology and Angiology, Medical Clinic I, University Hospital of the Justus Liebig University, Giessen, Germany.
- Department of Radiology, University Hospital of the Goethe University, Frankfurt/Main, Germany.
- Department of Cardiology, Asklepios Klinikum Harburg, Hamburg, Germany.
Abstract
Accurate pre-procedural computed tomography (CT) analysis is essential for optimal valve sizing and clinical outcomes in transcatheter aortic valve implantation (TAVI). Recently, fully automated, artificial intelligence (AI)-based CT analysis platforms have been developed to simplify and standardize this process. The aim of the study was to investigate the clinical impact of this new analysis method on the selection of valve prosthesis size. Overall, 247 patients with symptomatic severe aortic stenosis were enrolled. Patients underwent TAVI procedures at two different heart centres. The pre-procedural datasets were analysed by a standard TAVI CT-analysis software (3M, Pie Medical Imaging BV, The Netherlands) and a fully-automated CT-analysis-platform employing a deep-learning based algorithm. Key annular measurements and simulated prosthesis size selection were compared between both methods. The mean aortic annulus diameter was 24.5 ± 2.3 mm (3mensio) and 24.4 ± 2.4 mm (AI), respectively, with a mean absolute error (MAE) of 0.6 mm and mean absolute percentage error (MAPE) of 2.6%. Annulus perimeter (76.9 ± 7.0 mm vs. 74.6 ± 7.3 mm; MAE: 2.0 mm; MAPE: 2.6%) and annulus area (458.4 ± 87.2 mm<sup>2</sup> vs. 440.9 ± 85.6 mm<sup>2</sup>; MAE: 21.2 mm<sup>2</sup>; MAPE: 4.6%) showed excellent correlation (intraclass correlation coefficients > 0.95). Prosthesis size selection simulated on the basis of AI-derived measurements would have differed from the implanted size in 21% of patients, compared with 14% when using the semi-automated method. In this retrospective study, fully automated AI-based CT analysis demonstrated excellent agreement with conventional semi-automated measurements of the aortic annulus. Nevertheless, similar to established planning workflows, expert interpretation remains crucial to integrate the broader anatomical and clinical context required for optimal prosthesis size selection.