Investigation of the Impact of Cross-Frequency Coupling on the Assessment of Depression Severity through the Analysis of Resting State EEG Signals

Neurology Clinical Psychology Medical Imaging Medicine Psychiatry
cross-frequency coupling resting EEG signals major depressive disorder depression severity classification machine learning

Background

Depression, particularly Major Depressive Disorder (MDD), is a widespread and debilitating psychological disease often described as the “common cold” of mental health. Many people with MDD experience symptoms such as persistent sadness, hopelessness, cognitive impairment, and loss of motivation for daily activities, severely affecting personal and social life. Globally, the impact of depression is profound, affecting more than 340 million people to varying degrees. Moreover, the COVID-19 pandemic and its containment measures, such as social isolation and grief, have exacerbated the prevalence of depression. It is predicted that by 2030, depression will become the leading cause of disability, surpassing cardiovascular diseases, with annual deaths due to depression expected to reach one million. Given its high prevalence, disability rate, mortality rate, and relapse rate, timely detection and intervention of depression are particularly important.

Traditionally, the assessment of depression severity relies on clinical evaluations and interviews, such as the Beck Depression Inventory (BDI-II) and the Hamilton Depression Rating Scale (HAM-D). The accuracy of these assessments depends on the clinician’s experience and the reliability of the information provided by the patient, without the aid of biomarkers, which may lead to diagnostic errors.

In recent years, with the advancement of neuroimaging technologies such as functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and magnetoencephalography (MEG), researchers have shown great interest in objective biomarkers of depression. EEG, due to its cost-effectiveness, accessibility, and high temporal resolution, has been increasingly applied in studies of various neurological diseases, including depression. For instance, Li et al. decoded potential factors of Alzheimer’s disease using EEG and achieved a recognition accuracy of 98.10% with the Takagi-Sugeno-Kang classifier.

However, most existing studies on depression focus on single EEG frequency bands, using power spectrum analysis or applying functional connectivity standards (such as coherence and phase-locking between different brain regions) to assess the complexity of a particular brain area. These methods have limited effectiveness in detecting connectivity between different frequency bands, as neural processing in the brain likely depends on interactions between different frequency bands. This is where the concept of Cross-Frequency Coupling (CFC) comes in. CFC refers to the statistical correlation between activities in different frequency bands, which has been found to be of significant importance in various cognitive and perceptual processes as well as in disease states.

Source

This paper was published in the journal “Biomedical Signal Processing and Control” on May 17, 2024. The main authors are Parisa Raouf, Vahid Shalchyan, and Reza Rostami. Raouf and Shalchyan are affiliated with the Neuroscience and Neuroengineering Research Laboratory at the Iran University of Science and Technology, while Rostami is from the Department of Psychology and Educational Sciences at the University of Tehran.

Research Methods and Procedures

Research Method

This study explores the potential of four types of CFC (Phase-Amplitude Coupling PAC, Phase-Phase Coupling PPC, Frequency-Amplitude Coupling FAC, and Frequency-Frequency Coupling FFC) in identifying depression severity by analyzing resting-state EEG signals. The subjects included 22 depressed patients (12 with severe depression, 10 with moderate depression) and 15 healthy participants, analyzed using 19-channel EEG data. The study first calculated four types of CFC between low-frequency bands (Delta, Theta1, Theta2, Alpha1, Alpha2) and high-frequency bands (Beta and Gamma1), then selected 10 features showing significant differences and applied them to four classifiers: Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), K-nearest Neighbor (KNN), and Decision Tree (DT).

Detailed Procedure

Data Acquisition and Preprocessing

Resting-state EEG data were recorded for 7 minutes using a Mitsar-EEG 201 device, with electrode placement based on the international 10-20 system. The recorded brain signal frequency range was 0.5 to 70Hz, and the sampling frequency was 1000Hz. During data preprocessing, a 50Hz notch filter and a 0.5 to 45Hz FIR band-pass filter were used to effectively remove power line noise, and artifact removal was performed using independent component analysis (ICA) and the motion artifact removal algorithm (MARA) plugin. To ensure consistency and reduce bias from different recording lengths, all participants’ EEG signals were processed to a length of 5 minutes.

Feature Extraction

For comprehensive CFC evaluation, the study calculated instantaneous amplitude, instantaneous phase, and instantaneous frequency for each time sample using the Hilbert transform for each frequency band, with each time window being 6 seconds. Then, Kruskal-Wallis statistical tests were used to select features showing significant differences (p-value < 0.05).

Types of Cross-Frequency Coupling

  1. Phase-Amplitude Coupling (PAC): Describes how low-frequency phase changes affect high-frequency amplitude changes.
  2. Phase-Phase Coupling (PPC): Measures the phase synchronization between different oscillations.
  3. Frequency-Amplitude Coupling (FAC): Describes how frequency changes within one band affect the amplitude of signals in another band.
  4. Frequency-Frequency Coupling (FFC): Shows how frequency changes in one band are induced by frequency changes in another band.

Classification and Validation

Ten CFC features showing significant differences were chosen, and the performance of four classification models was evaluated using 5-fold cross-validation. The classifiers used were SVM, KNN (with k set to 10), LDA, and DT, with SVM classification performed using a one-vs-one method.

Results

Cross-Frequency Coupling at Each Electrode

Kruskal-Wallis statistical analyses revealed significant differences at the O2 electrode in PAC and FFC indicators, indicating the presence of more pronounced functional dysfunctions in the right hemisphere. Analysis of PAC indicators revealed that PAC values increased with increased depression severity, whereas FAC values were higher in the healthy group and lower in the severe depression group.

Cross-Frequency Coupling Between Electrodes

When analyzing CFC between multiple frequency bands across electrodes, the FFC indicators showed the most significant statistical differences in all calculations, with the main diagonal values in FFC feature calculations showing full correlation. Specifically, PPC between Theta1-Gamma1 bands in the right temporal lobe was most affected in the severe depression group.

Conclusion and Significance

This study reveals significant associations between depression severity and CFC features extracted from EEG signals. Among the extracted features, the maximum PPC was found to have the highest precision in classifying depression severity, reaching up to 91.43%. This indicates that cross-frequency coupling methods have significant potential in situational interventions and therapeutic efficacy evaluations.

Additionally, the study found that depression notably affects the right hemisphere’s temporal, parietal, and occipital regions. The results provide a theoretical basis for clinical decisions and personalized treatment, contributing to more accurate assessments of depression severity.

Highlights

  1. Innovative Method: This is the first systematic study on the effectiveness of four types of CFC in classifying depression severity.
  2. High Classification Accuracy: Achieved a maximum classification accuracy of 91.43%, which is significantly higher than traditional methods.
  3. Exploration of Neural Mechanisms: Revealed potential mechanisms of interaction between different brain regions and frequency bands in depression.

This study provides a new perspective and method for assessing depression severity in psychiatric clinical practice through cross-frequency coupling analysis. It lays a solid foundation for subsequent research and applications.