Deep Learning Analysis of LVEF Dynamics After Valve-in-Valve TAVI

K. Hug (München)¹, M. Tokodi (Budapest)², M. Lachmann (München)³, T. Trenkwalder (München)¹, F. Roski (München)¹, C. Pellegrini (München)¹, J. Hübner (München)¹, C. Salzmann (München)¹, F. Syryca (München)¹, M. Jurisic (München)¹, F. Haug (München)¹, H. A. Alvarez Covarrubias (München)¹, E. Xhepa (München)¹, A. Kastrati (München)¹, H. Schunkert (München)¹, M. Joner (München)¹, T. Rheude (München)^1
1Deutsches Herzzentrum München Klinik für Herz- und Kreislauferkrankungen München, Deutschland; ²Semmelweis University Heart and Vascular Center Budapest, Ungarn; ³Klinikum rechts der Isar der Technischen Universität München Klinik und Poliklinik für Innere Medizin I München, Deutschland

Background

Left ventricular ejection fraction (LVEF) often improves after transcatheter aortic valve implantation (TAVI), and early recovery of LVEF has been associated with improved long-term outcomes. Valve-in-Valve (ViV)-TAVI is increasingly used to treat degenerated bioprosthetic valves with favorable hemodynamic results; however, the evolution of LVEF after ViV-TAVI remains poorly characterized.

 

Aims

This study aims to evaluate the evolution of LVEF in patients with degenerated aortic bioprostheses undergoing ViV-TAVI using both conventional echocardiography and artificial intelligence.

 

Methods

Changes in LVEF from baseline to 3-month follow-up were retrospectively analyzed using data from a single-center German registry including 183 consecutive patients who underwent ViV-TAVI for degenerated aortic bioprostheses between 2017 and 2025. LVEF was assessed using both conventional methods and the QUEST-EF deep learning algorithm applied to two-dimensional apical four-chamber echocardiographic videos. Patients were further stratified according to the underlying mode of bioprosthetic valve failure (predominant intravalvular regurgitation (AR) vs. Restenosis (AS)).

 

Results

In a multifactorial analysis of variance, LVEF varied significantly over the 3-month follow-up period and between predominant AR and AS as the primary modes of bioprosthetic valve (p=0.011 F(2,180)=16.11, η_p²=0.152).

In patients with AR (n=78), LVEF showed a statistically significant transient decline immediately after the procedure (∆LVEF_{pre vs. post} -3.57 %; p<0.001), followed by a significant recovery at 3-month follow-up (∆LVEF_{post vs. 3MFU} +6.47%; p<0.001).

In contrast, patients with predominant AS (n=105) exhibited stable or slight increased LVEF values without an early post-procedural drop, which did not reach statistical significance.

 

Conclusion

Automated deep-learning–based assessment of LVEF revealed distinct temporal recovery patterns after ViV-TAVI. Patients with predominant aortic regurgitation experienced a transient post-procedural decline in systolic function, reaching levels comparable to those patients with aortic stenosis, suggesting reversible myocardial adaptation.