Analyzing SCG methodology for identification of ventricular depolarization events
Mousumi Das,Tilendra Choudhary,L.N. Sharma,M.K. Bhuyan
DOI: https://doi.org/10.1016/j.bspc.2024.106940
IF: 5.1
2024-10-25
Biomedical Signal Processing and Control
Abstract:Systolic time intervals (STIs), such as pre-ejection period (PEP), left ventricular ejection time (LVET) and PEP-to-LVET ratio, often measured by echocardiography, play vital roles in assessing cardiac contractility, cardiac efficiency, and sympathetic activity. However, to facilitate these measurements for wearables, multiple cardiac signals are needed, resulting in more required sensing devices. In this study, we developed a machine learning (ML) approach to estimate ventricular depolarization (VD) timings using only a single low-cost microelectromechanical system (MEMS)-based accelerometric sensor for measuring the STIs. A cardio-mechanical seismocardiogram (SCG) signal, manifested by rhythmic chest-vibrations, was acquired from the sensor. The SCG signal effectively captures the fiducial points required for measuring LVET. However, computing PEP involves VD timing instants, that are usually identified by the R-peaks of the electrocardiogram (ECG) signal. Consequently, measuring PEP requires more than one cardiac signal. To avoid the use of multiple sensing modalities, this study aims to compute R-peaks instants directly from the SCG signal. We devised a deep feedforward neural network (DFN) with SCG-based features to estimate VD events, which were referenced by R-peaks of concurrent electrocardiogram (ECG) signal. The estimated R-peaks along with the required SCG fiducial points can be used to measure the STIs. A set of self-recorded real-time SCG data collected in supine position was used to train and validate the model. The performance was validated using mean absolute error (MAE), root mean square error (RMSE), and standard deviation of error (SDE). Experimental results showed that the VD events detected using the proposed model has a MAE of 0.080, RMSE of 0.167, and SDE of 0.164. Also, the proposed DFN model was compared with classical ML models, such as random forest and eXtreme Gradient Boosting, where our DFN outperformed all. This demonstrates the standalone utilization of SCG as a cardiac tool for measuring left ventricular systolic functions, offering potential applications in assessing various ventricular anomalies.
engineering, biomedical