Reusable wearable PPG representations must capture both local pulse morphology and broader physiological regularities. Existing pre‑training objectives usually provide supervision at the whole‑recording or pair level, overlooking that different concepts become reliable only over specific waveform evidence scopes. To address this, we propose HiPPT, a physiology‑grounded framework that matches each supervision source to its appropriate evidence scope.
We build a PPG‑Text corpus containing over 4 million wrist‑worn recordings from 202 K users, pairing each recording with morphology descriptions derived from the waveform and context descriptions derived from metadata. HiPPT combines record‑level morphology alignment—grounding individual recordings in pulse characteristics—with cross‑recording context alignment—aggregating recordings that share the same metadata tuple to capture context‑consistent regularities. Waveform‑guided text modeling further recovers masked contextual concepts from PPG patch representations, providing fine‑grained waveform‑text supervision.
We evaluate on six downstream tasks covering blood pressure, vascular stiffness, physiological attributes, and cross‑dataset heart‑rate estimation. HiPPT achieves the best results on primary linear‑probing metrics. The aligned space also supports zero‑shot profiling of age group, sex, and BMI category, individually or jointly, with performance improving as more recordings are aggregated. These findings demonstrate that evidence‑scope matching yields transferable and semantically accessible wearable PPG representations. Code and pretrained weights will be released.
Review