A dual-branch EEG decoding method based on dynamic SHAP weight analysis

The dual-branch EEG decoding method based on dynamic SHAP weight analysis, combined with local and global feature extraction modules, dynamically calculates channel contribution weights, solving the problem of insufficient information mining in existing EEG diagnoses of major depressive disorder, and achieving higher recognition accuracy and interpretability.

CN121188546BActive Publication Date: 2026-06-26SOUTHWEST JIAOTONG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511369334.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-09-24
Publication Date
2026-06-26
Estimated Expiration
2045-09-24

AI Technical Summary

Technical Problem

Existing methods for diagnosing major depressive disorder using EEG rely on single features or isolated frameworks, failing to fully explore the multi-level and heterogeneous information in EEG, thus limiting the improvement of recognition performance. At the same time, SHAP, as a post-hoc interpretation tool, is not fully integrated into the model building process, resulting in a lack of local foundational support for global representation and insufficient model interpretability.

Method used

A dual-branch EEG decoding method based on dynamic SHAP weight analysis is adopted. By combining local feature extraction module and global feature extraction module with dilated causal convolution and multi-head self-attention model, the contribution weight of each channel is dynamically calculated, multi-scale features are fused and redundant information interference is reduced, thereby improving the accuracy and interpretability of the model.

Benefits of technology

Effective extraction of multi-scale EEG features reduces redundant information interference, captures various dependencies, improves the accuracy and interpretability of the model, and enhances its diagnostic ability for major depressive disorder.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121188546B_ABST
    Figure CN121188546B_ABST
Patent Text Reader

Abstract

The application discloses a kind of double-branch EEG decoding method and system based on dynamic SHAP weight analysis, comprising the following steps: after resting-state electroencephalogram signal is collected and preprocessed, it is mapped to two-dimensional space matrix;Through local feature extraction module, multi-scale local features are constructed and extracted, and local features are integrated into a unified multidimensional feature set, the feature vector after fusion is analyzed in depth by global feature extraction module, the complex dependence between features is captured and differentiated weight is given;During the global feature extraction process, the contribution weight of each channel to the classification performance is dynamically calculated using SHAP to improve the classification performance and interpretability of the model;Finally, the trained multi-head self-attention model decodes electroencephalogram, and features are extracted.By adopting double-branch parallel structure to extract inter-channel phase synchronization features and space-time dependence features, while combining SHAP to further improve the accuracy, reliability and interpretability of the model.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Motion imagination electroencephalogram classification method based on interpretable lightweight model

    CN118133096A

  • Stroke prediction multi-architecture stacked ensemble supermodel

    WO2023133427A1