Biosignal Fingerprinting: A Cross-Modal PPG-ECG Foundation Model

Zhangdaihong Liu, Chang Liu, Fenglin Liu, Yixuan Chen, Yang Yang†, David A. Clifton, Xiao Gu† (Oxford / SJTU / Oxford Suzhou)

Summary

Oxford/SJTU M2AE (Multi-modal Masked Autoencoder) learns compact biosignal fingerprints from >3.4M paired 10 s lead-II ECG + PPG segments (MIMIC-III WDB + VitalDB via PulseDB processing). Dual modality-specific Transformer encoders merge at a shared bottleneck with dual decoders; trained with reconstruction + cross-modal contrastive (same-subject cross-time/cross-modality positives). Fingerprints support frozen linear-probe / XGBoost downstream and cross-modal waveform reconstruction even when only one modality is present at inference — the wearable-relevant claim.