SIGMA-PPG: Statistical-prior Informed Generative Masking Architecture for PPG Foundation Model

TL;DR

SIGMA-PPG addresses the challenge of noise and redundancy in PPG foundation models by proposing a two-stage architecture that combines a spectrum-aware VQ-VAE tokenizer with semantic consistency, and a Prior-Guided Adversarial Masking scheme driven by statistical priors to shape a curriculum for a Transformer encoder. The model is pre-trained on over 120,000 hours of data and fine-tuned on 12 diverse downstream tasks, achieving state-of-the-art performance on many tasks, particularly in morphology-sensitive regression. Key contributions include the PSD-oriented spectral reconstruction objective, the semantic consistency constraint, and the priors-based RL masking via Gumbel-Top-k sampling, which together enable robust, physiology-aligned representations. The work demonstrates significant potential for a unified physiological encoder in clinical and wearable settings, while acknowledging domain shifts between clinical and ambulatory data and outlining paths toward on-device deployment and broader data coverage.

Abstract

Current foundation model for photoplethysmography (PPG) signals is challenged by the intrinsic redundancy and noise of the signal. Standard masked modeling often yields trivial solutions while contrastive methods lack morphological precision. To address these limitations, we propose a Statistical-prior Informed Generative Masking Architecture (SIGMA-PPG), a generative foundation model featuring a Prior-Guided Adversarial Masking mechanism, where a reinforcement learning-driven teacher leverages statistical priors to create challenging learning paths that prevent overfitting to noise. We also incorporate a semantic consistency constraint via vector quantization to ensure that physiologically identical waveforms (even those altered by recording artifacts or minor perturbations) map to shared indices. This enhances codebook semantic density and eliminates redundant feature structures. Pre-trained on over 120,000 hours of data, SIGMA-PPG achieves superior average performance compared to five state-of-the-art baselines across 12 diverse downstream tasks. The code is available at https://github.com/ZonghengGuo/SigmaPPG.

Figures (10)

• Figure 1: Overview of the SIGMA-PPG model framework. The architecture consists of two cascading stages: (1) Stage 1: Spectrum-Aware Semantic Tokenizer. At this stage, a VQ-VAE is used to map continuous raw PPG signals into discrete semantic tokens. A Power Spectral Density (PSD) reconstruction objective is employed to capture physiological frequency characteristics. Furthermore, a semantic consistency constraint is introduced by a vector quantization to ensure that physiologically identical waveforms, even those perturbed by artifacts or noise, map to consistent codebook indices. (2) Stage 2: Prior-Guided Masked Generative Pre-training. This stage employs a Reinforcement Learning-driven Teacher-Student framework to replace standard random masking. The Teacher network employs statistical priors (amplitude and skewness) to construct a dynamic curriculum and generate challenging masking policies, i.e., Prior-Guided Adversarial Masking. This mechanism effectively guides the Student network (a Transformer Encoder) to avoid noise overfitting and capture global morphological dependencies.
• Figure 2: Comparisons among different masking methods. The colors of the bars correspond to four distinct masking mechanisms. No.: random masking, i.e. a standard uniform random masking without any prior knowledge; +knowledge: static prior-based probabilistic masking, i.e. a heuristic approach where masking probability is statically determined by skewness and amplitude; +Teacher: unconstrained adversarial masking, i.e. dynamic masking generated by a Teacher without prior constraints; +knowledge+Teacher: the proposed Prior-Guided Adversarial Masking, where the Teacher efficiently targets physiologically significant structures under the guidance of statistical priors.
• Figure 3: Examples of different masking policies. From the top panel to the bottom: Random Masking, Unconstrained Adversarial Masking, Static Prior Masking, and the proposed Prior-Guided Adversarial Masking. Only a 30-second signal segment is displayed.
• Figure 4: Examples of how statistical scoring functions are used. Note how the score drops to near-zero during flat region ($t<20s$) but remains high ($\approx 1.0$) for clean, high-amplitude signals, effectively creating a "Safe Zone" for valid physiological data. For clarity's sake, just a 30-second signal segment is displayed. Colors provides indications of score values, from poor (blue) to high (red) quality signal.
• Figure 5: Impact of semantic consistency constraint. We compare the performance of SIGMA-PPG with (red) and without (grey) the consistency loss $\mathcal{L}_{Con}$ under Linear Probing and Full Fine-tuning protocols, the two graphs on the left and on the right, respectively. The lower MAE and the higher AUC demonstrate that $\mathcal{L}_{Con}$ significantly enhances representation robustness against noise in regression tasks and improves semantic separability in classification tasks, serving as a critical component for both frozen feature extraction and downstream adaptation.