Autocorrelation-based signal quality assessment for ballistocardiogram: a peak-free approach

Ballistocardiogram (BCG) signal quality assessment (SQA) is critical for reliable cardiac monitoring and disease prediction. Existing SQA methods for physiological signals typically rely on peak detection, which is unreliable for BCG due to its multi-lobed morphology. We propose a peak-free SQA method that evaluates signal periodicity through windowed autocorrelation, retains power spectral concentration only as an auxiliary descriptor, and aggregates the autocorrelation peaks by the median into a robust subject-level index. Evaluated on a 46-subject dataset with synchronized reference ECG, the method yields an interpretable quality stratification. Using the reference ECG as ground truth, the index is strongly predictive of BCG usefulness: high-quality windows achieve substantially lower heart-rate error and a much higher proportion of clinically usable segments compared to low-quality windows. In discriminating usable windows, the proposed index significantly outperforms peak-based and statistical SQI baselines. Furthermore, the screened 1D-ConvVAE pipeline achieved an AUC of 0.815 versus 0.675 for the full cohort. A same-model, within-cohort stratified evaluation showed higher AUC on good- than lower-quality subjects (0.765 vs.\ 0.587); a filter ablation indicates that bandpass filtering alone does not account for the observed difference. The method provides a reliable, peak-free SQA approach for BCG, enabling automatic quality screening for large-scale datasets.