A Signal Quality Index for Ballistocardiogram Recordings based on Electrocardiogram RR Intervals and Matched Filtering

Abstract

Detection of noisy ballistocardiogram (BCG) measurements with a matched filter is investigated. The BCG signal is segmented by the RR intervals of the concurrently recorded electrocardiogram (ECG). These segments are called BCG heartbeats and are averaged to estimate a parent template and sub-templates of BCG heartbeats leveraging the quasi-periodic nature of the heart. All BCG heartbeats in the measurement are used to estimate the parent template BCG heartbeat. By contrast, sub-templates are estimated using subsets of BCG heartbeats that extend over several breaths to capture the variations in the phase of the BCG heartbeat due to the breathing cycles. The parent template BCG heartbeat in a recording is regarded as the subject-specific matched filter. Finally, the average correlation coefficient between sub-templates and the parent template BCG heartbeat in a recording is proposed as the signal quality index (SQI). A seven-step experiment was performed among thirteen subjects. For reliable parent template estimation, at least 40 BCG heartbeats should be recorded. For detection of noisy BCG measurements, the area under the receiver operating characteristics curve was presented for conditions of motion artifacts, speaking, and deep breathing. Motion artifacts could be detected with lower false positive rates for the same constraint on the detection rate, when compared to speaking and deep breathing cases. This work may facilitate online detection of noisy BCG recordings.