PPG-LM: A Photoplethysmography–Language Model with Multi-Level Clinical Alignment

Xiaoda Wang, Minxiao Wang, Maxwell A Xu, Patrick Langer, Kaiqiao Han, Defu Cao, Xiao Luo, Yuzhe Yang, Yan Liu, Xiao Hu, Yizhou Sun, Wei Wang, Carl Yang

Summary

First claimed PPG–language model family spanning segment, medication-event, and visit levels. Automatic captions from signal measurements + structured EHR (MC-MED); two-stage CoCa-style contrastive+captioning then time-aware event/visit alignment (~73.7k h PPG pretrain). Beats LLM and PulseLM baselines on recognition/retrieval/captioning; Stage-2 frozen probes beat PaPaGei-S, Pulse-PPG, AnyPPG, SIGMA-PPG, MOMENT-L, Chronos-2 on reported clinical tasks.