A Novel Approach to Recognize Valvular Heart Diseases Based on Morphological Similarity of Heartbeats in Seismocardiography Signals

Valvular Heart Diseases (VHD) are cardiac pathologies involving heart valves malfunctioning, which are generally diagnosed via heart sounds auscultation or Phonocardiography (PCG). Auscultation requires a skilled clinician to listen to a subject’s chest via a stethoscope, so it is limited to patient visits in clinical settings. PCG recordings could be used for continuous, prolonged monitoring in non-clinical settings, but its susceptibility to environmental noises can severely impair the quality of recordings and the diagnostic reliability. Seismocardiography (SCG) is a cardio-mechanical monitoring technique that measures the weak accelerations of the chest-wall due to heartbeats, and captures the infrasonic vibrations produced by heart valves. In this study, it was observed that the heartbeats morphology varies significantly in SCG signals of patients with VHDs, as compared to healthy subjects. The normalized cross-correlation (NCC) was considered as a measure of morphological similarity, and was computed between all heartbeats detected in each SCG recording. The mean and standard deviation of NCC values were provided as features to several machine learning algorithms to classify VHD patients and healthy subjects. Gaussian Support Vector Machine and Linear Discriminant Analysis achieved the highest overall accuracy of 94.2%, with sensitivities and PPV of 97% and 96% for pathological subjects, and of 80% and 84% for healthy subjects. The promising yet preliminary results of this proof-of-concept study suggest that VHDs cause a morphological instability in SCG signals, which could provide new insights into the mechanical manifestations of such pathologies, e.g., to improve early diagnosis or timely recognition of exacerbations.