Scholarlink

Background <p>Myocardial bridging (MB) is a common congenital coronary anomaly that is increasingly recognized as a potential cause of myocardial ischemia and recurrent chest pain, even in the absence of obstructive coronary artery disease. However, reliable noninvasive risk stratification for recurrent chest pain in patients with MB remains challenging. Advanced coronary computed tomography angiography (CCTA)–derived biomarkers, including CT-based fractional flow reserve (CT-FFR) and the pericoronary fat attenuation index (FAI), may capture complementary functional and inflammatory mechanisms underlying MB-related symptoms, but their combined prognostic value has not been fully clarified.</p> Methods <p>Patients diagnosed with myocardial bridging on CCTA were consecutively enrolled. Clinical characteristics and CCTA-derived anatomical, functional, and inflammatory parameters were systematically collected. Recurrent chest pain during follow-up was defined based on standardized clinical criteria. After feature selection using Cox proportional hazards regression, a random survival forest (RSF) model was constructed to predict recurrent chest pain. Model performance was evaluated using concordance indices and time-dependent receiver operating characteristic analyses. Model interpretability was assessed using Shapley additive explanations (SHAP).</p> Results <p>A total of 174 patients with MB were included and 54 patients (31.03%) experienced recurrent chest pain during follow-up. Multivariable analysis identified advanced age, female sex, diastolic luminal stenosis of the bridged segment, systolic compression index, CT-FFR ≤ 0.80, elevated FAI and increased pericoronary adipose tissue (PCAT) volume as independent predictors of recurrent chest pain. SHAP-based interpretability analysis demonstrated that the predictive model achieved good discrimination, with area under the receiver operating characteristic curve (AUC) values of 0.895 (95% CI, 0.854–0.926) in the training cohort, 0.837 (95% CI, 0.781–0.889) in the validating cohort, and 0.804 (95% CI, 0.718–0.872) in the independent testing cohort.</p> Conclusions <p>An explainable machine learning model integrating CT-derived functional and inflammatory imaging biomarkers supports individualized prediction of recurrent chest pain in patients with myocardial bridging. This approach may improve noninvasive risk stratification and support personalized clinical decision-making.</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

Predicting recurrent chest pain in patients with myocardial bridging using explainable machine learning: a multicenter study integrating CT-derived functional and inflammatory imaging biomarkers

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月11日

查看全文
  • Jingquan Hu,
  • Kuigang He,
  • Jing Luo,
  • Rui Xia,
  • Yufeng Zhang,
  • Xiaolong Ma,
  • Zhihong Shao,
  • Xiaolong Gao

摘要

Background

Myocardial bridging (MB) is a common congenital coronary anomaly that is increasingly recognized as a potential cause of myocardial ischemia and recurrent chest pain, even in the absence of obstructive coronary artery disease. However, reliable noninvasive risk stratification for recurrent chest pain in patients with MB remains challenging. Advanced coronary computed tomography angiography (CCTA)–derived biomarkers, including CT-based fractional flow reserve (CT-FFR) and the pericoronary fat attenuation index (FAI), may capture complementary functional and inflammatory mechanisms underlying MB-related symptoms, but their combined prognostic value has not been fully clarified.

Methods

Patients diagnosed with myocardial bridging on CCTA were consecutively enrolled. Clinical characteristics and CCTA-derived anatomical, functional, and inflammatory parameters were systematically collected. Recurrent chest pain during follow-up was defined based on standardized clinical criteria. After feature selection using Cox proportional hazards regression, a random survival forest (RSF) model was constructed to predict recurrent chest pain. Model performance was evaluated using concordance indices and time-dependent receiver operating characteristic analyses. Model interpretability was assessed using Shapley additive explanations (SHAP).

Results

A total of 174 patients with MB were included and 54 patients (31.03%) experienced recurrent chest pain during follow-up. Multivariable analysis identified advanced age, female sex, diastolic luminal stenosis of the bridged segment, systolic compression index, CT-FFR ≤ 0.80, elevated FAI and increased pericoronary adipose tissue (PCAT) volume as independent predictors of recurrent chest pain. SHAP-based interpretability analysis demonstrated that the predictive model achieved good discrimination, with area under the receiver operating characteristic curve (AUC) values of 0.895 (95% CI, 0.854–0.926) in the training cohort, 0.837 (95% CI, 0.781–0.889) in the validating cohort, and 0.804 (95% CI, 0.718–0.872) in the independent testing cohort.

Conclusions

An explainable machine learning model integrating CT-derived functional and inflammatory imaging biomarkers supports individualized prediction of recurrent chest pain in patients with myocardial bridging. This approach may improve noninvasive risk stratification and support personalized clinical decision-making.