The paper introduces a stage-wise reconstruction framework for low-quality red and infrared PPG. It first pretrains an SpO₂ predictor on high-quality segments, then trains a masked reconstruction model with waveform loss, STFT-based frequency loss, and an SpO₂-preservation constraint. The reported subject-level MAE is 2.882% on the public OpenOximetry Repository and 2.359% on a private wearable PPG dataset. The approach targets physiologically relevant reconstruction rather than waveform similarity alone.
No heat snapshots are available in the last 24 hours.