Deep Learning on Whole-Body CT Predicts Short-Term Survival After TAVI

A deep learning model applied to pre-procedural whole-body CT predicted short-term survival (<12 months) after TAVI with AUC 0.787, sensitivity 85.3%, and specificity 72.1% in internal 5-fold cross-validation of 317 patients.

• The 3D DenseNet121 model was trained as a regression to predict survival in days, with binary short-term survival evaluation.
• Performance was comparable to state-of-the-art multimodal machine learning models requiring extensive clinical data, without manual feature engineering.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.