Deep Learning for Sleep Heart Rate Estimation from Accelerometers: Toward Population-Scale Cardiac Insight Without Optical Sensors
SeqSmoother estimates sleep heart rate from wrist accelerometry at 1.60 bpm participant-macro MAE.
SeqSmoother is a transformer temporal corrector that estimates sleep heart rate from wrist accelerometry using spectral descriptors, a Nightbeat frequency anchor, and a physics-motivated sub-harmonic feature. ECG is used only to build reference labels, not at inference. On 13 participant-disjoint folds it achieved a participant-macro MAE of 1.60 bpm. On matched intervals Nightbeat was more accurate (0.615 versus 1.091 bpm) but produced estimates for 72.85% of the eligible grid; the sub-harmonic ratio identified harmonic lock-on with AUROC 0.972.
- Transformer uses only wrist accelerometry features at inference time
- Participant-macro MAE is 1.60 bpm across 13 held-out folds
- Nightbeat is more accurate but covers 72.85% of the eligible grid
- Sub-harmonic ratio reaches AUROC 0.972 for harmonic lock-on
Full article204 words · extracted from arxiv.org · click to collapse
Large longitudinal cohorts often contain wrist accelerometry without optical heart-rate sensing, motivating recovery of cardiac information from motion signals already collected during sleep. We present SeqSmoother, a transformer-based temporal corrector for sleep heart rate (HR) estimation from wrist accelerometry. SeqSmoother combines spectral descriptors with an intermediate Nightbeat-derived frequency anchor and a physics-motivated sub-harmonic feature designed to identify harmonic frequency lock-on. All inference-time features are derived from wrist accelerometry, while ECG is used only to construct reference HR labels and training-label quality weights. We evaluate SeqSmoother using 13 participant-disjoint held-out folds and compare it with the official Nightbeat implementation under a matched 60-s window and 15-s step protocol. Across all out-of-fold predictions, SeqSmoother achieved a participant-macro MAE of 1.60 bpm. On Nightbeat-retained matched intervals, Nightbeat achieved lower absolute error than SeqSmoother (0.615 versus 1.091 bpm), while SeqSmoother provided estimates over a larger portion of the eligible recording; Nightbeat produced final estimates for 72.85% of the SeqSmoother-eligible out-of-fold grid. Separately, the proposed sub-harmonic ratio achieved an AUROC of 0.972 for identifying reference-defined harmonic lock-on candidates. These findings reveal an accuracy-availability trade-off between learned temporal modeling and quality-gated signal processing while providing empirical support for a physics-informed approach to identifying frequency-tracking failures in accelerometer-based sleep HR estimation.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.06823