Machine Learning Tool Identifies Coronary Microvascular Disease in Mice

A machine learning neural network named CORAL identified coronary microvascular disease in mice with 81 percent to 100 percent accuracy using non-invasive coronary blood flow patterns acquired through transthoracic Doppler echocardiography, according to research published in Cardiovascular Diabetology.

Direct diagnosis of coronary microvascular disease in human patients has historically remained elusive, requiring clinicians to rely on combinatorial criteria that incorporate both invasive and non-invasive procedures. Prior experimental data in mice indicated that coronary blood flow patterns contain characteristic fingerprints capable of serving as novel non-invasive diagnostic indicators.

To evaluate these blood flow signatures, investigators tested whether machine learning of Doppler coronary blood flow patterns could directly and accurately identify coronary microvascular disease.

The investigators compiled comprehensive in vivo datasets comprising approximately 500 mice across different stages of disease progression. Primary training and testing cohorts included control mice and diabetic db/db mice evaluated at early stages at 12 weeks of age, established disease stages at 16 weeks of age, and senescent stages at 36 weeks of age, totaling 460 animals. An independent test cohort evaluated an additional 40 mice subjected to a western-diet-fed model of coronary microvascular disease.

All participating animals were evaluated for coronary blood flow and cardiac function using echocardiography alongside invasive pressure-volume loop catheterization. Resulting coronary blood flow images were processed and supplied to Coronary Risk Assessment with Learning, designated as CORAL, a classifier designed with the optional ability to incorporate cardiac function features through a multilayer perceptron architecture.

When evaluated solely on coronary blood flow pattern analysis, CORAL achieved an accuracy of 100 percent at 12 weeks of age, detecting early-stage coronary microvascular disease in the animal cohort. For established coronary microvascular disease, the algorithm correctly identified the condition with 81 percent accuracy in the internal test dataset and 83 percent accuracy in the independent test dataset.

Adding semi-simultaneous measurements of non-invasive and invasive cardiac function improved the performance of the flow-based classifier. With cardiac functional features included, diagnostic accuracy rose to 89 percent in the test dataset and reached 90 percent in the independent test dataset, with accuracy reaching up to 100 percent depending on the timing of disease evaluation.

The authors concluded that CORAL reveals previously unutilized features within coronary blood flow patterns to identify coronary microvascular disease early in its progression in animal models. The researchers noted that translational application to clinical practice remains prospective, envisioning that further refinements to the algorithm could eventually enable direct and accurate diagnosis of coronary microvascular disease in human patients.

The accepted peer-reviewed manuscript was published ahead of the final Version of Record in Cardiovascular Diabetology under an open-access license.

Subject of Research: Machine learning detection of coronary microvascular disease in mouse models

Article Title: CORAL: a machine learning algorithm to identify coronary microvascular disease from doppler coronary blood flow patterns

Article References: Trask, A. J., Ueyama, Y., Bossenbroek, J., McDermott, M. R., McCallinhart, P. E., Erdengasileng, A., Sunyecz, I. L., Ray, W. C., & Bartlett, C. W. (2026). CORAL: a machine learning algorithm to identify coronary microvascular disease from doppler coronary blood flow patterns. Cardiovascular Diabetology. https://doi.org/10.1186/s12933-026-03396-6

Image Credits: AI Generated

DOI: 10.1186/s12933-026-03396-6

Keywords: coronary microvascular disease, machine learning, CORAL, Doppler echocardiography, coronary blood flow, cardiac function, mouse models, db/db mice, machine, learning, algorithm, identify