Foundation Models for Physiological Signals: Opportunities and Challenges

Simon A. Lee, Kai Akamatsu

Abstract: This review synthesizes the paradigm shift toward foundation models for analyzing physiological signals from wearable devices. We chart the landscape from the unique characteristics of wearable data—including photoplethysmography (PPG), electrocardiography (ECG), and accelerometry (ACC)—to the architectural and pre-training innovations that enable these models. We present a comprehensive taxonomy of prominent models analyzing their contributions to tasks like health monitoring, activity recognition, and disease prediction. The core of this review is a multi-faceted examination of the critical challenges and opportunities in this emerging research field.