PPGFlowECG: Latent Rectified Flow with Cross-Modal Encoding for PPG-Guided ECG Generation and Cardiovascular Disease Detection

TL;DR

PPGFlowECG addresses the gap between scalable PPG-based monitoring and diagnostic-grade ECG by learning a semantically aligned cross-modal latent space and performing ECG generation via latent rectified flow. The method integrates a CardioAlign Encoder to tightly couple PPG and ECG representations with semantic decodability, and a conditional latent flow to map Gaussian noise to ECG latents for accurate waveform synthesis. Extensive experiments on four datasets show improved translation fidelity and enhanced downstream CVD detection, including robust multi-label disease classification and AF detection, with strong zero-shot generalization and clinician-supported realism. The approach enables wearable-first cardiovascular screening by providing high-quality synthesized ECGs when standard ECG is unavailable, reducing diagnostic gaps in continuous monitoring pipelines.

Abstract

Electrocardiography (ECG) is the clinical gold standard for cardiovascular disease (CVD) assessment, yet continuous monitoring is constrained by the need for dedicated hardware and trained personnel. Photoplethysmography (PPG) is ubiquitous in wearable devices and readily scalable, but it lacks electrophysiological specificity, limiting diagnostic reliability. While generative methods aim to translate PPG into clinically useful ECG signals, existing approaches are limited by the misalignment of physiological semantics in generative models and the complexity of modeling in high-dimensional signals. To address these limitations, we propose PPGFlowECG, a two-stage framework that aligns PPG and ECG in a shared latent space using the CardioAlign Encoder and then synthesizes ECGs with latent rectified flow. We further provide a formal analysis of this coupling, showing that the CardioAlign Encoder is necessary to guarantee stable and semantically consistent ECG synthesis under our formulation. Extensive experiments on four datasets demonstrate improved synthesis fidelity and downstream diagnostic utility. These results indicate that PPGFlowECG supports scalable, wearable-first CVD screening when standard ECG acquisition is unavailable.

Figures (7)

• Figure 1: Top: Ventricular electrical activation on the ECG precedes and electromechanically initiates cardiac contraction, resulting in a delayed peripheral blood-volume pulse reflected in the PPG waveform. Bottom: PPG is easy to acquire but lacks diagnostic fidelity, whereas ECG reveals definitive disease markers. AI-based PPG-to-ECG translation offers a promising diagnostic pathway.
• Figure 2: Illustration of the proposed PPGFlowECG framework for high-fidelity PPG-to-ECG translation and cardiovascular disease detection. In this figure, the framework aligns PPG and ECG in a shared latent space using the CardioAlign Encoder and employs latent rectified flow to synthesize ECGs with high fidelity and interpretability.
• Figure 3: (a) A scatter plot comparing heart rate (HR) estimation accuracy and (b) An evaluation encompassing cardiovascular diseases across rhythm, conduction, vascular, and structural domains.
• Figure 4: 3D t-SNE visualizations.
• Figure 5: Ablation study of the effect of sampling steps.