Multimodal electromechanical phenotyping enables non-invasive classification of cardiac dysfunction

Background Cardiac dysfunction is characterised by coordinated electromechanical disturbances. Synchronized acquisition of electrocardiography (ECG), phonocardiography (PCG), impedance cardiography (ICG), and seismocardiography (SCG) enables simultaneous assessment of electrical, mechanical, and haemodynamic domains. Whether a single multimodal acquisition can classify echocardiographic abnormalities across distinct functional domains remains unclear. Objective To evaluate whether integrated multimodal feature sets discriminate echocardiographic abnormalities across ventricular, atrial, diastolic, and valvular domains, and whether performance improves when controls are restricted to a low-comorbidity subgroup, supporting target-specific rather than non-specific cardiovascular signatures. Methods A total of 1069 participants from a prospective single-centre observational cohort underwent synchronized ECG/PCG/ICG/SCG acquisition, standard ECG, and comprehensive transthoracic echocardiography. A multimodal feature space was constructed spanning electrical, haemodynamic, and cross-modal coupling domains. Target echocardiographic findings included LV systolic dysfunction, RV strain, tricuspid annular plane systolic excursion (TAPSE) ≤17 mm, severe left atrial (LA) enlargement, advanced diastolic dysfunction, and tricuspid regurgitation. For each target, logistic regression models were trained and cross-validated on the integrated feature set. Models were evaluated in the full cohort and with controls restricted to a predefined low-comorbidity subgroup (excluding ischaemic heart disease, hypertension, diabetes, and other major comorbidities). Results In the full cohort, multimodal models discriminated all six targets (AUC 0.72 – 0.80) despite low prevalences of 6 – 13 % and a high level of comorbity. Restricting the controls to the low-comorbidity subgroup improved AUC of all targerts to 0.83 – 0.85. Performance gains were target-dependent, with the largest improvement observed for TAPSE ≤17 mm (+ 0.14; 0.72 to 0.85) and the smallest for tricuspid regurgitation (+ 0.07). Conclusion Synchronized multimodal electromechanical acquisition enables non-invasive classification of echocardiographic abnormalities across six functional domains. Improved discrimination after restricting controls to a low-comorbidity subgroup supports the presence of target-specific rather than merely non-specific cardiovascular signatures. These findings support prospective validation of multimodal electromechanical phenotyping for non-imaging triage and early detection of cardiac dysfunction.