Cross-Dataset Validation of a Sensor Agnostic Seismocardiography Peak Detection Method

Seismocardiography (SCG) is a non-invasive technique for capturing the mechanical vibrations of the heart, typically measured using inertial measurement units (IMUs) placed directly on the sternum, and offering insights into cardiac function beyond traditional electrocardiography. In this study, a novel SCG peak detection method is proposed that involves a multi-step rule-based algorithm. These detected peaks are then compared to the true R-peaks detected from time-synchronized ECG signals.The method proposed in this study was evaluated using five publicly available datasets, N = 235 subjects. The heterogeneous nature of these datasets allowed for assessment of the proposed peak detection algorithm under varying signal quality conditions and clinical scenarios. Results indicate that the proposed method demonstrates high precision in peak detection across presumed healthy subject datasets with different sensor set ups and measurement protocols, with a mean sensitivity (TPR), precision (PPV), and interbeat interval root mean square error (RMSE) of 96.46%, 97.42%, and 40.64 ms for healthy subjects. Performance shows a notable decline in subjects with valvular heart disease: TPR, PPV, RMSE: 86.17%, 92.73%, and 121.7 ms and in subjects where measurements were taken during a right heart catheter procedure TPR, PPV, RMSE: 71.39%, 79.65%, and 183.1 ms.This is the first such study using five open access SCG datasets together for validation of a SCG based peak detection method. All code and results have been open-sourced and are available on GitHub to foster reproducibility and further research in the community.CCS CONCEPTS• Applied computing ➝ Life and medical sciences ➝ Health informatics