Rethinking EEG-Based Disease Diagnosis: Decoupling Instance Representation Learning from Subject-Level Supervision

2026-07-29

AI Summary

This study addresses a critical limitation in conventional EEG-based disease diagnosis, where subject-level labels are naively assigned to all EEG segments, erroneously assuming uniform diagnostic value across instances and struggling to learn effective representations when the number of subjects is far smaller than that of instances. To overcome this, the authors propose BridgeMIL, a two-stage framework: first, a self-supervised pretraining phase leverages temporal proximity alignment and sub-bag sampling without instance-level labels; second, the pretrained encoder is transferred to an attention-based multiple instance learning (MIL) aggregator with a feature-preserving constraint to mitigate representation drift. This work is the first to decouple instance-level representation learning from subject-level supervision, thereby avoiding label inheritance bias. It further introduces a variance–covariance regularization that prevents representation collapse without requiring negative samples and reveals substantial heterogeneity in instance diagnostic reliability. Evaluated across three EEG datasets and five backbone architectures in 15 experiments, BridgeMIL achieves state-of-the-art performance in 14 cases, with an average accuracy of 76.57%—a 4.28 percentage point improvement over the strongest baseline—and yields more discriminative subject-level representation spaces.

Abstract

EEG-based disease diagnosis requires one prediction per subject, yet common pipelines segment recordings into short instances, inherit the subject label for every instance, and train instance-level classifiers. This assumes that all instances provide equally reliable diagnostic evidence. Multiple instance learning (MIL) avoids inherited labels by treating each subject as a bag. However, EEG datasets contain far fewer subjects than instances, which can limit the quality of the representations learned by end-to-end MIL. We propose BridgeMIL, a two-stage framework that decouples instance representation learning from subject-level supervision. Stage 1 pretrains the encoder without inherited instance labels by aligning temporally nearby windows and independently sampled within-subject sub-bags. Variance and covariance regularization prevent collapse and reduce redundancy without negative pairs. Stage 2 transfers the encoder to an attention-based MIL aggregator, applies supervision only to subject predictions, and limits representation drift through feature retention. Across three EEG disease datasets and five representative backbones, BridgeMIL attains the highest mean accuracy in 14 of 15 dataset-backbone settings and an overall mean accuracy of 76.57%, 4.28 percentage points higher than the strongest baseline. Further analyses reveal substantial variation in inherited-label reliability across instances, greater performance sensitivity to subject scarcity than to instance scarcity, and a more structured representation space with distinct subject-wise clusters and improved separation between diagnostic classes. Together, these findings underscore the importance of aligning supervision with the subject-level prediction objective while learning from abundant EEG instances without assigning disease labels to individual instances.

Problem

• EEG-based disease diagnosis
• instance-level labeling
• subject-level supervision
• multiple instance learning
• label reliability

Innovation

• BridgeMIL
• multiple instance learning
• EEG disease diagnosis
• self-supervised representation learning
• subject-level supervision