Continuous oxygen saturation (SpO$_2$) estimation from wearable photoplethysmography (PPG) is crucial for long-term health monitoring. However, low-quality red and infrared PPG segments can distort waveform morphology and degrade SpO$_2$ prediction accuracy. Existing PPG denoising and reconstruction methods typically focus on either waveform fidelity or heart rate characteristics, while relying solely on time-domain waveform loss is insufficient to preserve frequency structure and SpO$_2$-relevant information.
This paper proposes a SpO$_2$ predictor-guided stage-wise time-frequency reconstruction framework for low-quality dual-wavelength PPG signals. The proposed method first selects high-quality PPG segments to pretrain an SpO$_2$ predictor. A masked reconstruction model is then trained to recover randomly masked PPG regions using a joint reconstruction objective that combines time-domain waveform loss with frequency-domain loss computed from the short-time Fourier transform (STFT).
To ensure the reconstruction task is physiologically relevant, the pretrained SpO$_2$ predictor is incorporated as an additional constraint, encouraging the reconstructed PPG to preserve SpO$_2$ information rather than merely minimizing reconstruction error. The SpO$_2$ predictor and PPG reconstructor model are optimized through four training stages. Experiments on the public OpenOximetry Repository and a private wearable PPG dataset demonstrate that the proposed approach achieves the lowest subject-level MAE, with 2.882% on the public dataset and 2.359% on the private dataset.
Blogger's Review: This study innovatively addresses the reconstruction of low-quality PPG signals by integrating an SpO$_2$ predictor, significantly enhancing the accuracy of oxygen saturation estimation. This approach offers new insights for the further development of wearable health monitoring devices, particularly in signal processing and physiological information extraction.