From Brain to Heart: the novel cardioconnectomics paradigm

Myocardial infraction (MI), commonly known as heart attack, remains the leading cause of human death. The analysis of temporal features derived from 12-lead electrocardiographic (ECG) signals represents the mainstay for the diagnosis of cardiovascular diseases. However, MI-related abnormalities arise from impaired electrical activation of the cardiac muscle, leading to alterations in the global synchronization of cardiac activity. Here, we introduce, for the first time, the concept of cardioconnectomics, describing the heart as a complex network of interconnected myocardial regions, and apply network graph theory (NGT) to characterise the topological organisation of electrical synchronisation pattern. We further investigate the feasibility of integrating NGT-derived metrics within machine learning (ML) models to discriminate between healthy subjects and MI patients. The 12-lead ECG signals of 109 demographically balanced subjects for each group of interest (healthy controls (HCs) vs. MI patients) were retained from publicly available electrocardiography dataset (i.e., PTB-XL). The cardiac connectome was constructed by using each ECG lead as a cardiac area and the crosscorrelation coefficient as a measure of synchronization between each pair of leads. Six global NGT metrics were calculated for all subjects, i.e., global efficiency (GE), characteristic path length (CPL), clustering coefficient (CC), betweenness (BET), strength (STR) and eccentricity (ECC). Wilcoxon signed-rank test was used to compare topological properties of the two groups. Four ML models were implemented to assess which one provided the best discrimination performance (HCs vs MI) based on the NGT metrics. The MI connectomes showed lower efficiency (reduced GE, p-value = 2.032 x 10-10), less clustered (reduced CC, p-value = 7.806 x 10-9 ), and less coordinated electrical pathways (reduced BET, p-value = 1.653 x 10-12), with a globally reduced level of synchronization (reduced STR, pvalue = 3.848 x 10-10). The logistic regression model showed the best classifier performances (accuracy = 0.773). Overall, these findings may establish the NGT-derived metrics as a valuable alternative to conventional ECG parameters, ensuring high ML performance in discriminating HCs and MI patients and contributing to a deeper understanding of MI-induced cardiac alterations.