Explainable EEG-based prediction of depression therapy outcomes using local Fibonacci pattern analysis.

Area of Science:

• Neuroscience
• Computational psychiatry
• Biomedical engineering

Background:

• Predicting depression treatment outcomes for selective serotonin reuptake inhibitors (SSRI) and repetitive transcranial magnetic stimulation (rTMS) is challenging.
• Current trial-and-error approaches delay effective clinical benefit for patients with depression.

Purpose of the Study:

• To develop and validate an electroencephalogram (EEG)-based predictive model for depression treatment response.
• To utilize local Fibonacci pattern (LFP) features extracted from pre-treatment EEG signals to classify responders to SSRI and rTMS therapies.

Main Methods:

• Pre-treatment EEG data from SSRI and rTMS treatment groups were analyzed.
• Local Fibonacci Pattern (LFP) features were extracted from 19-channel EEG recordings.
• A feedforward neural network was employed for classification, with validation using 10-fold cross-validation and leave-one-subject-out (LOSO) methods.

Main Results:

• Segment-level 10-fold cross-validation yielded high accuracies (up to 100%) for predicting treatment responders.
• Subject-wise LOSO validation demonstrated more modest accuracies (61.83% for SSRI, 77.93% for rTMS), indicating potential overfitting.
• Discriminative EEG channels were identified in frontal, temporal, and parietal regions crucial for emotional and cognitive processing.

Conclusions:

• Local Fibonacci Patterns (LFP) offer a computationally efficient method for EEG-based depression treatment prediction.
• The model's performance in LOSO validation suggests a need for larger, independent datasets before clinical deployment.
• Further research with diverse cohorts is essential to validate the generalizability and clinical utility of this predictive approach.