Comparing resting state and task-based EEG using machine learning to predict vulnerability to depression in a non-clinical population

Kaushik, Pallavi; Yang, Hang; Roy, Partha Pratim; van Vugt, Marieke · 2023 · Crossref

DOI: 10.1038/s41598-023-34298-2

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of predicting vulnerability to Major Depressive Disorder (MDD) in non-clinical populations by comparing the efficacy of resting-state EEG (rs-EEG) and task-based EEG data. While MDD imposes a significant societal burden, existing research often relies on either rs-EEG or task-based data exclusively, leaving their relative predictive capabilities unclear. The authors specifically investigate whether EEG signals can capture "stickiness"—the difficulty in disengaging from negative thoughts, a key mechanism in rumination and depression relapse. By focusing on individuals with high versus low vulnerability rather than clinical patients, the study aims to identify sensitive biomarkers for early intervention. The researchers recruited 40 participants, stratified into high and low vulnerability groups based on standardized scores from the Perseverative Thinking Questionnaire, Rumination Response Scale, and Center for Epidemiologic Studies Depression Scale. EEG data were collected using a 32-channel Biosemi system. Participants underwent a 5-minute resting-state recording with eyes closed and a Sustained Attention to Response Task (SART). The SART included thought probes to measure the "stickiness" of thoughts, with EEG epochs corresponding to these trials forming the task-based dataset. Data were pre-processed using band-pass filtering, artifact removal via Independent Component Analysis, and baseline correction. To predict vulnerability, the authors employed machine learning models, including MultiLayer Perceptron, Decision Trees, 1D-Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM networks. Additionally, evolutionary algorithms (Grey Wolf Optimization, Genetic Algorithm, Particle Swarm Optimization) were used to identify the most informative subset of rs-EEG biomarkers. The results revealed distinct neural patterns associated with depression vulnerability. In rs-EEG, individuals with high vulnerability exhibited increased amplitude in left frontal channels and decreased amplitude in right frontal and occipital channels. In task-based EEG, high vulnerability was associated with increased amplitude in right temporal, occipital, and parietal regions, whereas low vulnerability correlated with increased central brain amplitude. Regarding predictive performance, the 1D-CNN achieved the highest accuracy (98.06%) using raw rs-EEG data, while the LSTM model achieved 91.42% accuracy using delta waves from task-based data. Feature selection via evolutionary algorithms identified Higuchi fractal dimension, phase lag index, correlation, and coherence as the most critical rs-EEG features for classification. The study concludes that rs-EEG is superior for predicting depression vulnerability due to higher classification accuracy and ease of collection. However, task-based EEG provides valuable insights into the cognitive mechanisms driving depression, such as rumination and thought stickiness. These findings suggest that while rs-EEG is optimal for diagnostic screening, task-based paradigms remain essential for understanding the underlying neural processes of depressive symptoms. The integration of machine learning with specific EEG biomarkers offers promising avenues for future non-invasive diagnostics and early detection tools.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices