A comprehensive review of biosignal foundation models

Abstract

Foundation models are emerging as a promising approach to biosignal analysis, offering solutions to common challenges in learning from biosignal data, such as high variability, noise or limited annotations. This survey provides a comprehensive review of the state-of-the-art in this rapidly advancing field. We introduce commonly used biosignal modalities (including EEG, ECG, EMG, EOG and PPG) and discuss their defining characteristics with respect to training large models. We present a structured survey of existing work across unimodal and multimodal settings covering data processing, feature extraction, model architectures, pre-training paradigms and evaluation methods. We identify open challenges such as benchmarking, interpretability and data availability and highlight possible directions for future research. By consolidating our findings from the current literature and opportunities for progress, we aim to guide the development of robust and universal foundation models for biosignals.

The authors' abstract, as published at the source. Journal of Neural Engineering, 2026 · DOI ↗