EEG-MACS: Manifold Attention and Confidence Stratification for EEG-based Cross-Center Brain Disease Diagnosis under Unreliable Annotations

Zhenxi Song, Ruihan Qin, Huixia Ren, Zhen Liang, Yi Guo, Min zhang, Zhiguo Zhang

Abstract: Cross-center data heterogeneity and annotation unreliability significantly challenge the intelligent diagnosis of diseases using brain signals. A notable example is the EEG-based diagnosis of neurodegenerative diseases, which features subtler abnormal neural dynamics typically observed in small-group settings. To advance this area, in this work, we introduce a transferable framework employing **M**anifold **A**ttention and **C**onfidence **S**tratification (**MACS**) to diagnose neurodegenerative disorders based on EEG signals sourced from four centers with unreliable annotations. The MACS framework’s effectiveness stems from these features: 1) The _**Augmentor**_ generates various EEG-represented brain variants to enrich the data space; 2) The _**Switcher**_ enhances the feature space for trusted samples and reduces overfitting on incorrectly labeled samples; 3) The _**Encoder**_ uses the Riemannian manifold and Euclidean metrics to capture spatiotemporal variations and dynamic synchronization in EEG; 4) The _**Projector**_, equipped with dual heads, monitors consistency across multiple brain variants and ensures diagnostic accuracy; 5) The _**Stratifier**_ adaptively stratifies learned samples by confidence levels throughout the training process; 6) Forward and backpropagation in **MACS** are constrained by confidence stratification to stabilize the learning system amid unreliable annotations. Our subject-independent experiments, conducted on both neurocognitive and movement disorders using cross-center corpora, have demonstrated superior performance compared to existing related algorithms. This work not only improves EEG-based diagnostics for cross-center and small-setting brain diseases but also offers insights into extending MACS techniques to other data analyses, tackling data heterogeneity and annotation unreliability in multimedia and multimodal content understanding. We have released our code here: https://anonymous.4open.science/r/EEG-Disease-MACS-0B4A.

Primary Subject Area: [Engagement] Emotional and Social Signals

Secondary Subject Area: [Content] Media Interpretation, [Experience] Interactions and Quality of Experience, [Generation] Multimedia Foundation Models

Relevance To Conference:

- This work introduces MACS, a framework that employs Manifold Attention and Confidence Stratification to address data heterogeneity and annotation unreliability in EEG modeling—a challenge that resonates across various topics in the analysis and interpretation of multimedia and multimodal data.

- Particularly, our proposed framework has been validated on both cognitive and movement neurodegenerative diseases using cross-center data, which exacerbates the challenges of subtler abnormal neural dynamics and small-group settings. Additionally, the theme 'Emotional and Social Signals' focuses on analyzing emotional, cognitive (e.g., brain-based), and interactive social behavior from individual to small group contexts.

- Our framework's methodology and the accompanying open-source code address critical challenges in multimedia and multimodal processing. By integrating manifold space with Euclidean metrics, we provide novel perspectives on EEG representation learning that are applicable to emotional, cognitive, and related research areas. Additionally, our confidence stratification-based learning mechanism can be adapted to facilitate other low-resource multimedia representation learning projects.

Submission Number: 5285