Research on Depression Recognition Model and Its Temporal Characteristics Based on Multiscale Entropy of EEG Signals

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) is a cost-effective tool for depression diagnosis.
• Current EEG depression models often overlook information from coarser temporal scales.
• Multiscale analysis offers a novel approach to extract richer temporal characteristics from EEG signals.

Purpose of the Study:

• To investigate the feasibility of multiscale analysis for enhancing depression recognition models.
• To explore the temporal characteristics of EEG signals at different scales for depression detection.
• To optimize machine learning models for depression diagnosis using multiscale entropy.

Main Methods:

• Utilized two types of multiscale entropy to analyze EEG signals.
• Developed machine learning models using Linear Discriminant Analysis (LDA), Logistic Regression (LR), Radial Basis Function Support Vector Machine (RBF-SVM), and K-Nearest Neighbors (KNN).
• Performed mathematical analysis to examine the relationship between temporal scale and model performance.

Main Results:

• Achieved a maximum classification accuracy of 96.42% using the KNN classifier at scale 3.
• Scales 3 and 9 demonstrated superior performance across various classifiers compared to other scales.
• Model performance showed a correlation with scale variation, suggesting an optimal scale exists within a finite range.

Conclusions:

• Multiscale analysis is a practical and effective method for building and optimizing EEG-based depression recognition models.
• An optimal temporal scale can significantly improve model performance with predictable computational costs.
• Further research into scale-dependent model capabilities can advance computer-assisted diagnosis for depression.