Survival Prediction of TAVI Patients Using End-to-End Deep Image Regression on Whole-Body CT.

INTRODUCTION

Transcatheter aortic valve implantation (TAVI) is a standard treatment for patients with severe aortic stenosis. However, predicting long-term survival outcomes remains challenging, as traditional risk scores such as EuroSCORE II and STS-PROM exhibit poor calibration for TAVI cohorts.

METHODS

In this work, we developed an end-to-end deep learning pipeline to predict overall survival of TAVI patients directly from pre-procedural whole-body CT scans. The pipeline uses a 3D DenseNet121 architecture within the AUCMEDI framework, trained as a regression model to predict survival in days, using 5-fold cross-validation on 317 TAVI patients.

RESULTS

The model achieved an AUC of 0.787, a specificity of 72.1%, and a sensitivity of 85.3% for identifying short-term survival patients (<12 months) in a binary evaluation, comparable to state-of-the-art multimodal ML models requiring extensive clinical data.

CONCLUSION

End-to-end deep learning on routinely acquired CT scans can effectively identify high-risk TAVI patients, offering potential as a clinical decision support tool for pre-operative assessment without manual feature engineering.