Scholarlink

Abstract <p>Coronary calcification is a prevalent pathology and strong cardiovascular indicator. However, its progression drivers remain poorly defined. With limited clinical samples, it is unclear if simple traditional machine learning can effectively predict progression, challenging risk assessment and individualized treatment. We assembled a serial CCTA dataset of 2,579 patients from West China Hospital. Using Random Forest, Gradient Boosting Decision Trees, XGBoost, and Logistic Regression with SHAP analysis, we identified key features of coronary artery calcification progression and built predictive models, then compared them with traditional clinical models. The Random Forest (RF) model achieved an AUC of 0.81 (95% CI: 0.78–0.84) vs. 0.64 (0.59–0.68) for the traditional model. Baseline CACS, plaque burden, and other CCTA features were key predictors, with critical thresholds determined. A coronary artery calcification progression prediction score (CACPPS) was derived to quantify personalized progression risk. The RF model also showed acceptable calibration (Brier score = 0.169; Hosmer-Lemeshow p = 0.415), favorable decision-curve net benefit, and an optimal CACPPS threshold of 0.566. In patients with suspected or confirmed CAD, a traditional yet interpretable machine learning model can predict CACS progression and generate CACPPS to support treatment decisions and dynamic individualized management, mitigating black-box concerns through indirect interpretability.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction and critical feature analysis for coronary artery calcification progression

  • 期刊论文

  • 发表日期：

    2026年8月3日

查看全文
  • Ran Liu,
  • Gaojian Yang,
  • Wenyu Huang,
  • Junyan Zhang,
  • Yuting Lei,
  • Rui Zhang,
  • Zhongxiu Chen,
  • Yong He,
  • Hongmei Yan,
  • Kaiyue Diao

摘要

Abstract

Coronary calcification is a prevalent pathology and strong cardiovascular indicator. However, its progression drivers remain poorly defined. With limited clinical samples, it is unclear if simple traditional machine learning can effectively predict progression, challenging risk assessment and individualized treatment. We assembled a serial CCTA dataset of 2,579 patients from West China Hospital. Using Random Forest, Gradient Boosting Decision Trees, XGBoost, and Logistic Regression with SHAP analysis, we identified key features of coronary artery calcification progression and built predictive models, then compared them with traditional clinical models. The Random Forest (RF) model achieved an AUC of 0.81 (95% CI: 0.78–0.84) vs. 0.64 (0.59–0.68) for the traditional model. Baseline CACS, plaque burden, and other CCTA features were key predictors, with critical thresholds determined. A coronary artery calcification progression prediction score (CACPPS) was derived to quantify personalized progression risk. The RF model also showed acceptable calibration (Brier score = 0.169; Hosmer-Lemeshow p = 0.415), favorable decision-curve net benefit, and an optimal CACPPS threshold of 0.566. In patients with suspected or confirmed CAD, a traditional yet interpretable machine learning model can predict CACS progression and generate CACPPS to support treatment decisions and dynamic individualized management, mitigating black-box concerns through indirect interpretability.

Graphical abstract