Magnetocardiography as a noninvasive diagnostic strategy for suspected coronary microvascular dysfunction

Background

Coronary microvascular dysfunction (CMD) is increasingly recognized as an independent predictor of mortality with a 4-fold higher risk. However current diagnostic modalities are limited by the need for an invasive procedure, access, cost, and exposure of ionizing radiation.

Purpose

To investigate the ability of magnetocardiography (MCG) to identify CMD in patients with suspected ischemia and no obstructive coronary artery disease (INOCA).

Methods

This is an observational, prospective pilot study of patients scheduled for coronary functional angiography (CFA), gold standard for evaluation for CMD (defined as coronary flow reserve (CFR) ≤2). 13 patients underwent both CFA and a noninvasive 36-channel MCG scan. A machine learning model was developed to characterize the presence of CMD in these patients against age matched controls (AMCs). The model consists of a logistic regression classifier which takes features representing the relative strengths of the “characteristic dipoles” of the MCG scan as input. The characteristic dipoles are parameterizations of the three strongest magnetic field map components resulting from a singular value decomposition of the MCG signal. A total of 37 patients were included in this analysis including 13 patients who completed CFA (7 had CMD and 6 had CFR >2 and included in the AMCs group). An additional 24 asymptomatic, healthy patients that did not undergo CFR were also included in the AMC group.

Results

The mean age for AMCs was 57 years (70% women) and mean age for CMD patients was 54 years (100% women). The performance of the model was evaluated using repeated stratified cross-validation with 5 folds and 3 repeats, resulting in 15 different 80%/20% train/test splits. Figure 1 shows the distribution of samples belonging to the CMD and AMC groups in a 2-dimensional representation of the feature space. The clear separation of the two groups and the clustering of the AMCs demonstrates the ability of the model to identify patients with CMD. We found that MCG had a mean accuracy of 94.8% (±6.4%), sensitivity of 100% (±0.0%) and specificity of 93.3% (±8.2%) for the detection of CMD using gold standard CFR ≤2 as reference.

Conclusion(s)

First study to show that MCG can be used with 94.8% accuracy to identify CMD among patients suspicious for INOCA with no exposure to ionization, 90 seconds of scan time and minimal cost. The use of this noninvasive modality to identify CMD warrants further investigation.

Funding Acknowledgement

Type of funding sources: Private company. Main funding source(s): Genetesis