Single-center study tests MRI-based machine learning for artery-level revascularization prediction

A multivariable XGBoost model combining clinical variables with conventional cardiac magnetic resonance measurements, T2* mapping and rest myocardial blood flow predicted which major coronary arteries underwent revascularization during hospitalization in patients with stable coronary artery disease. In a retrospective single-center study of 843 patients, the model achieved receiver operating characteristic AUCs of 0.87 in the development training cohort and 0.94 in each of two reported validation sets. The findings support further evaluation of the approach for artery-level treatment prediction, while prospective multicenter validation remains necessary before clinical implementation.

The study included patients with stable coronary artery disease recruited between 2022 and 2025. Of the 843 patients, 468 formed the development cohort, and 375 patients recruited during a later, non-overlapping period formed a same-center temporal validation cohort. In the development cohort, 723 of 2,808 major coronary-artery observations underwent revascularization during hospitalization, representing 25.7% of the assessed arteries. Because the outcome was recorded for individual coronary arteries, all artery observations from the same patient were assigned to the same fold during cross-validation, reducing the possibility that information from one patient could be used to evaluate predictions for that same patient.

Researchers evaluated four machine-learning algorithms with six-fold patient-grouped cross-validation. Least absolute shrinkage and selection operator analysis identified four cardiac magnetic resonance features and three clinical variables for the model. The multivariable XGBoost model achieved an AUC of 0.871, compared with 0.815 for T2* mapping alone and 0.822 for rest myocardial blood flow alone; both comparisons were statistically significant, with p-values below 0.05. In separate comparisons, models using T2* mapping alone and rest myocardial blood flow alone achieved AUCs of 0.84 and 0.86, respectively, versus 0.87 for the combined model. These results show that the integrated multivariable model discriminated better than the individual CMR parameters and the reported single-parameter machine-learning models in the development data.

The model’s performance was also evaluated with area under the precision-recall curve. It reached precision-recall AUCs of 0.72 in the development training cohort, 0.90 in validation set 1 and 0.76 in validation set 2. The corresponding receiver operating characteristic AUCs were 0.87, 0.94 and 0.94 across the three sets. The study therefore evaluated artery-level in-hospital revascularization as a prediction task, rather than testing whether revascularization improved later clinical outcomes. It also did not establish that the model caused a different treatment decision or demonstrate its effect on patient health after discharge.

Several features of the evidence limit how broadly the findings can be interpreted. The study was retrospective and conducted at one center, and the temporal validation cohort was drawn from that same institution rather than from multiple independent centers. The analysis also focused on patients with stable coronary artery disease and on major coronary-artery observations, so the reported performance should not automatically be extended to other populations, coronary segments or clinical settings. Prospective multicenter testing would be needed to assess whether the model retains its discrimination when used across different hospitals and patient groups.

Subject of Research: Artery-level revascularization prediction in stable coronary artery disease using cardiac MRI and machine learning

Article Title: A machine learning model integrating T2* mapping and rest myocardial blood flow sequences for predicting revascularization of individual coronary arteries in stable coronary artery disease

Article References: Zhao, L., An, D.-A., Shi, R.-Y., Wu, R., Gu, Z.-Y., Chen, B.-H., Xue, S., & Wu, L.-M. (2026). A machine learning model integrating T2* mapping and rest myocardial blood flow sequences for predicting revascularization of individual coronary arteries in stable coronary artery disease. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02860-9

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02860-9

Keywords: machine learning, XGBoost, coronary artery disease, cardiac magnetic resonance, T2* mapping, myocardial blood flow, revascularization, retrospective validation, clinical prediction, machine, learning, model