Seismocardiography (SCG), a non-invasive method for capturing cardio-mechanical signals, is often susceptible to noise and motion artifacts. Current approaches primarily use automated algorithms and machine learning techniques for signal quality indexing and fiducial point detection. However, validation is hindered by the scarcity of standardized, high-quality annotated datasets. To facilitate the manual annotation process of SCG signals, we developed an open-source graphical user interface, providing a simple, efficient, and accurate tool. Using data from a porcine hypovolemia protocol, we annotated SCG waveforms for signal quality scores and annotated cardiac timing intervals (e.g., aortic opening and aortic closing) with 17,059 SCG heart beats in total across six porcine subjects. Validating the rigor and consistency of annotations, performed analyses for inter-annotator agreements resulted high agreements achieving strong correlations against gold-standard catheter-based measurements for aortic opening (AO) (r = 0.926) and aortic closing (AC) (r = 0.911). This expertly annotated dataset supports advancements in real-time cardiac monitoring, denoising, and diagnostic applications, and enhances reproducibility and comparability in SCG research across the biomedical signal processing community.
Cho et al. (Wed,) studied this question.