Key result
Novel transformer-based algorithm MS-BPT estimates heart rate with a superior ~1.2 bpm error.
Why the study?
Accurate heart rate estimation from ballistocardiography during real-world sleep is challenging due to non-stationary signals, motion interference, and unreliable supervision.
Does the Multi-Scale Ballistocardiography Pyramid Transformer (MS-BPT) algorithm improve the accuracy of heart rate estimation from ballistocardiography during sleep compared to other deep learning methods?
Does the Multi-Scale Ballistocardiography Pyramid Transformer (MS-BPT) algorithm improve the accuracy of heart rate estimation from ballistocardiography during sleep compared to other deep learning methods?
The MS-BPT model provides accurate and consistent heart rate estimation from ballistocardiography under real-world sleep conditions, outperforming existing deep learning methods.
May facilitate contactless HR monitoring in sleep; leaves open prospective clinical validation.
Accurate heart rate (HR) estimation from ballistocardiography (BCG) remains a challenge in real-world sleep monitoring because of non-stationary signals, motion interference, and unreliable supervision. To address these issues, we propose a Multi-Scale Ballistocardiography Pyramid Transformer (MS-BPT) for robust single-channel BCG-based HR estimation. MS-BPT integrates Pyramid Input (PI), Atrous Spatial Pyramid Pooling (ASPP), a Transformer Encoder Block (TEB), and an RR-variability-guided quality-aware weighting strategy within a unified framework to capture multi-resolution temporal information, multi-scale contextual representations, and long-range dependencies. The model was evaluated on an overnight BCG-ECG dataset using subject-wise five-fold cross-validation. MS-BPT achieved a Mean Absolute Error (MAE) of 1.23 bpm, Standard Deviation of Absolute Error (SDAE) of 1.83 bpm, and Pearson Correlation Coefficient (PCC) of 0.93. Across the five evaluated deep learning methods, the corresponding metric ranges were 1.45–2.40 bpm, 1.90–2.26 bpm, and 0.54–0.80, respectively. These results demonstrate that MS-BPT provides accurate and consistent HR estimation under real-world sleep conditions.
No takes yet. Share an insight, caveat, or question.
Sheng et al. (2026) studied Real-world sleep monitoring. Multi-Scale Ballistocardiography Pyramid Transformer (MS-BPT) vs. Other deep learning methods was evaluated on Mean Absolute Error (MAE) for heart rate estimation. The Multi-Scale Ballistocardiography Pyramid Transformer (MS-BPT) achieved accurate heart rate estimation with a Mean Absolute Error of 1.23 bpm compared to 1.45–2.40 bpm for other evaluated methods.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: