Time-Frequency Based Detection of Mitral Regurgitation Using Seismocardiogram Signal

Effective and accessible screening is vital for mitigating the progression of mitral regurgitation (MR). This study proposes a time-frequency approach for MR detection using seismocardiogram signals. The initial step involves motion artifact removal through a threshold-based exclusion method using root-mean-square (RMS) filtering. Artifact-free 3-second segments are then obtained. These segments undergo hilbert vibration decomposition and stationary wavelet transform decomposition to derive monocomponents and subbands, respectively. Novel features, including relative wavelet energy, log energy entropy, and multi-scale kurtosis, are extracted from each component. These features are input to a random forest (RF) classifier for MR detection. The model achieved an impressive accuracy of 97.60%, sensitivity of 98.22%, and specificity of 96.98% on a cohort of 13 MR patients and 20 healthy individuals. Our proposed method for MR detection showcases comparable results with existing approaches for various cardiac disorders. This preliminary work underscores the potential of SCG signals as a cost-effective and accessible diagnostic tool for early MR screening in larger populations.