Accurate and consistent automatic seismocardiogram annotation without concurrent ECG

Abstract

Seismocardiography (SCG) is the measurement of vibrations in the sternum caused by the beating of the heart. Precise cardiac mechanical timings that are easily obtained from SCG are critically dependent on accurate identification of fiducial points. So far, SCG annotation has relied on concurrent ECG measurements. An algorithm capable of annotating SCG without the use any other concurrent measurement was designed. We subjected 18 participants to graded lower body negative pressure. We collected ECG and SCG, obtained R peaks from the former, and annotated the latter by hand, using these identified peaks. We also annotated the SCG automatically. We compared the isovolumic moment timings obtained by hand to those obtained using our algorithm. Mean ± confidence interval of the percentage of accurately annotated cardiac cycles were [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] for levels of negative pressure 0, -20, -30, -40, and -50 mmHg. LF/HF ratios, the relative power of low-frequency variations to high-frequency variations in heart beat intervals, obtained from isovolumic moments were also compared to those obtained from R peaks. The mean differences ± confidence interval were [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] for increasing levels of negative pressure. The accuracy and consistency of the algorithm enables the use of SCG as a stand-alone heart monitoring tool in healthy individuals at rest, and could serve as a basis for an eventual application in pathological cases.

Contactless Mapping of Thoracic and Abdominal Movements: Applications for Seismocardiography

“…SCG signals from ICARE and accelerometer are both bandpass filtered at 15 -60 Hz and the three chosen fiducial points are identified automatically using previously described software [12].…”

Detection and Analysis of Heartbeats in Seismocardiogram Signals

“…The authors in [39], instead, study the performance of SCG annotation during a lower body negative pressure test in lying position: 94% coverage is achieved over approximately 21,600 heartbeats, with an annotation time error of 9 ± 9 ms (mean±standard deviation). In a similar setting [40], an RMSE of 40, 71, 26, 51, and 27 ms for increasing levels of lower body negative pressure; in turn, the average sensitivity in IM detection is 97.2%, 93.0%, 76.9%, 61.6%, and 65.0%, respectively. Moreover, the authors in [41] annotate heartbeats of subjects lying in bed for approximately 10 min: by fusing accelerometer and gyroscope information, they achieve high accuracy and precision scores (99.9% and 99.6%, respectively).…”

Contactless seismocardiography via deep learning radars

“…To do so, it implements an automatic labeling function for SCG recordings. It is worth noting that past work has made efforts to automatically label SCG recordings (albeit those obtained from on-body accelerometers) [25,29]; however, these past methods require additional on-body ECGs leads [25] or have demonstrated high error margins [29], making them undesirable for practical/medical use.…”