ML model predicts coronary calcium progression on serial CCTA

On serial coronary CT angiography, a random forest model predicted coronary artery calcification progression with AUC 0.81 (95% CI 0.78-0.84), outperforming a traditional model (AUC 0.64, 95% CI 0.59-0.68). Baseline calcium score and plaque burden were key predictors.

• The model was trained on 2,579 patients from a single center using Random Forest, Gradient Boosting, XGBoost, and Logistic Regression with SHAP analysis.
• A coronary artery calcification progression prediction score (CACPPS) was derived, with an optimal threshold of 0.566 and acceptable calibration (Brier score 0.169).

Automated summary

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