eCAPRI: a novel tool combining clinical and imaging data for post-TAVI mortality prediction.

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Transcatheter aortic valve implantation (TAVI) is a key treatment for calcified aortic stenosis (CAS), but predicting patient outcomes remains challenging.
• The existing CAPRI score uses manual thoracic aortic calcium (TAC) measurements and clinical factors, showing promise but requiring improvement for broader application.

Purpose of the Study:

• To develop and validate an enhanced CAPRI score (eCAPRI) that automates TAC volume measurement using deep learning.
• To integrate additional automatically extracted imaging biomarkers into the eCAPRI score for improved prediction of one-year mortality in TAVI patients.

Main Methods:

• A deep learning model was developed for automated TAC volume segmentation.
• The eCAPRI score incorporated automatically computed TAC volume and additional imaging biomarkers (ventricle volume, pulmonary artery diameter, abdominal muscle surface).
• The eCAPRI score was trained and validated against the original CAPRI score and EuroSCORE Logistic using AUC and calibration curves.

Main Results:

• Automated TAC segmentation achieved a Dice score of 0.777.
• The eCAPRI score demonstrated superior predictive performance with an AUC of 0.731, significantly outperforming CAPRI (AUC=0.669) and EuroSCORE Logistic (AUC=0.588).
• eCAPRI showed improved calibration compared to the other scores.

Conclusions:

• The eCAPRI score, leveraging automated TAC quantification and novel imaging biomarkers, significantly enhances one-year mortality prediction in TAVI candidates.
• This fully automated approach offers a standardized and more accurate tool for risk stratification, aiding clinical decision-making in TAVI procedures.