Joint Attention Mechanism Learning to Facilitate Opto-physiological Monitoring during Physical Activity

TL;DR

The paper addresses motion artefact contamination in PPG signals during physical activity by introducing AM-GAN, an attention-based GAN that fuses PPG with synchronized tri-axial acceleration to produce artefact-free PPG waveforms. The generator (1D U‑Net) uses a multi-sensor cross-attention mechanism to identify MA components, while a discriminator enforces realism against reference signals, with a reconstruction loss guiding accurate MA removal. Across four protocols and multiple datasets, AM-GAN delivers robust HR, RR, and SpO$_2$ estimates, showing strong intra- and cross-dataset performance and outperforming several state-of-the-art MA-removal methods. The work demonstrates practical potential for wearable opto-physiological monitoring, offering high fidelity waveform reconstruction and improved vital sign estimation under varying activity intensities, with future prospects for edge deployment and broader physiological integration.

Abstract

Opto-physiological monitoring including photoplethysmography (PPG) provides non-invasive cardiac and respiratory measurements, yet motion artefacts (MAs) during physical activity degrade its signal quality and downstream estimation concurrently. An attention-mechanism-based generative adversarial network (AM-GAN) was proposed to model motion artefacts and mitigate their impact on raw PPG signals. The AM-GAN learns how to transform motion-affected PPG into artefact-reduced waveforms to align with triaxial acceleration signals corresponding to artefact components gained from a triaxial accelerometer. The AM-GAN has been validated across four experimental protocols with 43 participants performing activities from low to high intensity (6--12km/h). With the public datasets, the AM-GAN achieves mean absolute error (MAE) for heart rate (HR) of 1.81 beats/min on IEEE-SPC and 3.86 beats/min on PPGDalia. On the in-house LU dataset, it shows the MAEs < 1.37 beats/min for HR and 2.49 breaths/min for respiratory rate (RR). A further in-house C2 dataset with three oxygen levels (16%, 18%, and 21%) was applied in the AM-GAN to attain a MAE of 1.65% for SpO2. The outcome demonstrates that the AM-GAN offers a robust and reliable physiological estimation under various intensities of physical activity.

Figures (12)

• Figure 1: The proposed AM-GAN comprising a joint attention mechanism together with a GAN.
• Figure 2: The generator network aligned with the attention mechanism. In the cross-attention mechanism, the noisy PPG (P) feature map is set as the Query tensor, while the combined feature maps for the triaxial ACC and VEL are set as Key and Value tensors.
• Figure 3: The discriminator structure.
• Figure 4: A representative example for the waveform comparison of the original noisy signal, the reference signal and the generated signal at different physical activity intensities; i.e., (a) at rest, (b) low-intensity activity (6$km/h$), (c) medium-intensity activity (9$km/h$), and (d) high-intensity activity (12$km/h$).
• Figure 5: Waveform quality validation on LU subjects (F-female, M-male, T-treadmill, C-cycling). (a) Waveform quality ($R$) for F01-treadmill. (b)Mean waveform quality ($\overline{\langle R \rangle}$) across all LU testing subjects.