EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Clinical-to-Wearable PPG Translation via Conditional Latent Diffusion: Bridging Finger-Infrared and Wrist-Green Measurements

Hyeon-Hwa Choi, Hyo-Chang Seo

IEEE Access · 2026 · 14 · 168365-168379

DOI: 10.1109/access.2026.3746233 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hyeon-Hwa Choi und 1 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: K-Brain Project of the National Research Foundation of Korea (NRF) funded by the Korean government, Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Photoplethysmography (PPG) is widely used for non-invasive cardiovascular monitoring, but a substantial domain gap remains between clinically acquired finger-infrared PPG (Finger-IR PPG) and wearable wrist-green PPG (Wrist-Green PPG). Finger-IR PPG is often available in large clinical repositories with rich physiological measurements or diagnostic annotations, whereas Wrist-Green PPG is widely used in consumer wearables but is typically noisier and more sparsely labeled. This gap limits the reuse of clinical PPG resources for wearable-healthcare applications. To address this problem, we propose a unidirectional Finger-IR-to-Wrist-Green PPG translation framework combining Transformer-based variational autoencoders with a conditional latent diffusion model. The framework learns domain-specific latent representations and models the paired cross-domain mapping in the target Wrist-Green latent space. The method was evaluated on paired Finger-IR/Wrist-Green test sets and further examined on an external multiclass arrhythmia dataset spanning six rhythm categories. In the best configuration on the paired internal test set, the proposed approach achieved a Pearson correlation coefficient of 0.893 and a heart-rate mean absolute error of 1.39 beats per minute. Ablation analyses showed that synthesis depended on temporally structured Finger-IR conditioning rather than only on target-domain signal priors. External arrhythmia analysis suggested partial retention of source-derived rhythm and interval information, although waveform-level pathological validity could not be directly verified because paired Wrist-Green references were unavailable. These findings support the feasibility of using clinical Finger-IR PPG to synthesize wearable-domain Wrist-Green PPG and provide a proof-of-concept for leveraging clinical Finger-IR resources in wearable-domain PPG modeling.

Förderung

  • K-Brain Project of the National Research Foundation of Korea (NRF) funded by the Korean government
  • Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government

Zitieren

APA

Choi, H., & Seo, H. (2026). Clinical-to-Wearable PPG Translation via Conditional Latent Diffusion: Bridging Finger-Infrared and Wrist-Green Measurements. IEEE Access, 14, 168365-168379. https://doi.org/10.1109/access.2026.3746233

Vancouver

Choi H, Seo H. Clinical-to-Wearable PPG Translation via Conditional Latent Diffusion: Bridging Finger-Infrared and Wrist-Green Measurements. IEEE Access. 2026;14:168365-168379. doi:10.1109/access.2026.3746233

BibTeX
@article{Choi2026,
  author = {Choi, Hyeon-Hwa and Seo, Hyo-Chang},
  title = {Clinical-to-Wearable PPG Translation via Conditional Latent Diffusion: Bridging Finger-Infrared and Wrist-Green Measurements},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {168365--168379},
  doi = {10.1109/access.2026.3746233}
}
Kostenlos starten →