RESUMO
PURPOSE: Coronary microvascular dysfunction (CMD) is emerging as an important cause of myocardial ischemia, but there is a lack of a non-invasive method for reliable early detection of CMD. AIM: To develop an electrocardiogram (ECG)-based machine learning algorithm for CMD detection that will lay the groundwork for patient-specific non-invasive early detection of CMD. METHODS: Vectorcardiography (VCG) was calculated from each 10-second ECG of CMD patients and healthy controls. Sample entropy (SampEn), approximate entropy (ApEn), and complexity index (CI) derived from multiscale entropy were extracted from ST-T segments of each lead in ECGs and VCGs. The most effective entropy subset was determined using the sequential backward selection algorithm under the intra-patient and inter-patient schemes, separately. Then, the corresponding optimal model was selected from eight machine learning models for each entropy feature based on five-fold cross-validations. Finally, the classification performance of SampEn-based, ApEn-based, and CI-based models was comprehensively evaluated and tested on a testing dataset to investigate the best one under each scheme. RESULTS: ApEn-based SVM model was validated as the optimal one under the intra-patient scheme, with all testing evaluation metrics over 0.8. Similarly, ApEn-based SVM model was selected as the best one under the intra-patient scheme, with major evaluation metrics over 0.8. CONCLUSIONS: Entropies derived from ECGs and VCGs can effectively detect CMD under both intra-patient and inter-patient schemes. Our proposed models may provide the possibility of an ECG-based tool for non-invasive detection of CMD.
