PPG-Bench: A Unified Benchmark for PPG Foundation Models
Arvasu Kulkarni, Mahir Jain, Aditya Ray Mishra, Sandeep Singh, Siddharth Panwar (Mannas AI)
Summary
Independent open-source benchmark of five public PPG FMs (AnyPPG, PaPaGei-S, PaPaGei-P, PulsePPG, SigmaPPG) across nine open-access datasets and 26 tasks (17 primary + 9 secondary), under four downstream regimes (linear probe, random forest, frozen MLP, end-to-end finetune) with model-specific HPO and subject-disjoint CV. Controls: random/constant baseline, deterministic PRV feature set, and generic time-series FM MOMENT. Headline: 35.3% of 68 task–regime combos show no meaningful FM gain over random/constant; in 83.8% no PPG-specific FM beats both PRV and MOMENT.
Key facts
- License
- Not stated
Reported benchmarks
Cuffless blood pressure
ppg-bp · MAE (mmHg) ↓
PPG-Bench linear probe SBP/DBP MAE; AnyPPG; subject-level; subject-disjoint 5-fold
Heart rate estimation
ppg-bp · MAE (bpm) ↓
PPG-Bench linear probe; AnyPPG; subject-level aggregation; subject-disjoint 5-fold
Heart rate estimation
PPG-DaLiA · MAE (bpm) ↓
PPG-Bench linear probe; AnyPPG; subject-disjoint 5-fold; mean±95% CI in paper
Heart rate estimation
WildPPG · MAE (bpm) ↓
PPG-Bench linear probe; AnyPPG; subject-disjoint 5-fold
Updated: Oct 12, 2026
Related Research
- Cuffless blood pressure leaderboard
Benchmark results for cuffless blood pressure
- Heart rate estimation leaderboard
Benchmark results for heart rate estimation
- LIMA-Mamba: A lightweight and distribution-robust Mamba network for PPG-based biometrics
Research paper
- A Comparative Study of Signal Representations Methods and Deep Learning Architectures for PPG-Based Cuffless Blood Pressure Estimation
Seven 1D→2D transforms × several CNN variants on the UCI BP database. Best: CWT, errors reported as
- A Decodability-Grounded Audit of Clinical Prediction Claims in PPG Foundation Models
Audits frozen representations of four encoders on 37,784 MC-MED encounters via concept decodability,