Detecting multiple fiducial markers from a camera seismocardiogram

Seismocardiogram (SCG) records the cardiac contractions and valve functions, holding a promising avenue for the early detection of adverse cardiovascular events. To monitor such events, it is crucial to accurately identify the fiducial markers in the SCG signal in order to support the estimation of hemodynamic parameters. In this paper, we propose a novel framework that consists of a defocused camera-based speckle imaging system to measure the SCG signal and a deep learning based model (DiAT) to detect multiple fiducial markers. Specifically, DiAT employs dilated convolution and a multi-head attention mechanism to capture the temporal context of SCG signals for detecting seven fiducial markers (MC, IM, AO, IC, RE, AC, MO). The percentage of predicted values falling in the range of true values ± 1 ms, and the Mean Absolute Error (MAE) between the predicted value and the true value are used as evaluation metrics. Experiments involving 17 adult subjects with ice water stimulation protocol show that DiAT obtains 87.78% accuracy and 8.4106 ms MAE. It demonstrates that our proposed method has reliable performance in detecting seven fiducial markers of camera-SCG signals.