Depression recognition from EEG based on nonlinear analysis and adaptive feature fusion.

Area of Science:

• Neuroscience
• Affective Computing
• Machine Learning

Background:

• Machine learning shows promise for electroencephalograph (EEG)-based depression recognition.
• Existing methods struggle to fuse heterogeneous features effectively, limiting performance.

Purpose of the Study:

• To develop a novel affective computing framework for depression identification using EEG.
• To enhance depression recognition by integrating nonlinear analysis and adaptive feature coupling.

Main Methods:

• Utilized diverse entropy measures to analyze complex dynamics in EEG signals.
• Implemented a weighted average fusion strategy for adaptive aggregation of multi-view features.
• Mitigated redundant feature influence through the proposed fusion strategy.

Main Results:

• The proposed framework significantly improved the accuracy of depression recognition on public datasets.
• EEG entropy values were lower in patients with depression compared to healthy controls, indicating reduced brain activity complexity.
• Visualization analysis supported the findings on altered brain complexity.

Conclusions:

• Nonlinear analysis and adaptive feature fusion offer valuable insights for EEG-based depression recognition.
• The framework demonstrates the potential for improved diagnostic accuracy in mental health through advanced signal processing.
• Reduced EEG complexity in depression patients highlights potential biomarkers for the condition.