SPOTR: Spatio-temporal Pooling One-Token Reconstruction for Universal Physiological Signal Self-supervised Learning

2026-06-20

AI Summary

Existing self-supervised methods exhibit limited performance across heterogeneous physiological signals, often compromising clinically relevant structures or relying on redundant information, which results in suboptimal linear probing accuracy and high computational overhead. This work proposes a novel compression–reconstruction pretraining framework featuring a single-token global bottleneck mechanism. The approach employs a spatiotemporal compression module to condense entire multimodal physiological recordings—including EEG, iEEG, ECG, and PPG—into a single global token, from which the original signals are reconstructed exclusively. By preserving essential physiological structures while drastically reducing computational and memory demands, the method achieves substantial gains: under linear probing, average AUC improves by 4.64%–21.71%, inference latency decreases by 78%, and peak GPU memory usage drops by 52% compared to general-purpose time-series foundation models.

Abstract

Physiological signals such as EEG, ECG, and PPG are widely used in clinical monitoring. Recent self-supervised learning (SSL) methods offer an attractive way to leverage unlabeled recordings, yet they still fall short in practice. In particular, current SSL methods struggle across heterogeneous datasets, often distorting clinically meaningful structures or learning shortcuts from temporal and cross-channel redundancy. Consequently, existing SSL methods often deliver limited performance under linear probing, a lightweight adaptation setting that better matches real-world medical scenarios. Moreover, most Transformer-based SSL models encode a flattened spatiotemporal token sequence, incurring high computation and memory cost, and are typically developed within a single modality. To address these limitations, we present SPOTR (Spatio-temporal Pooling One-Token Reconstruction), a compress-reconstruct pretraining framework that introduces a single-token global bottleneck for physiological signals. SPOTR compresses each waveform into a single-token representation and reconstructs the signal conditioned only on this representation. Meanwhile, SPOTR introduces an efficient spatio-temporal compaction module to reduce computation and memory cost. Pretrained on 20 datasets spanning EEG, iEEG, ECG, and PPG, SPOTR consistently outperforms the strongest baseline under linear probing, improving average AUC by 18.49%, 21.71%, 17.86%, and 4.64%, respectively. Compared with a representative general-purpose time-series foundation model, SPOTR achieves around 78% lower latency and 52% lower peak GPU memory on average. The code can be found at https://github.com/5GYYYYY/SPOTR.

Problem

• physiological signals
• self-supervised learning
• heterogeneous datasets
• linear probing
• spatio-temporal redundancy

Innovation

• one-token reconstruction
• spatio-temporal pooling
• self-supervised learning
• physiological signals
• cross-modal pretraining