Deep Learning-Based Artificial Intelligence Can Differentiate Treatment-Resistant and Responsive Depression Cases...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Treatment-resistant depression (TRD) affects many patients, necessitating efficient identification methods.
• Current depression diagnosis and treatment outcome prediction often utilize electroencephalogram (EEG) data.
• No prior studies have applied deep learning (DL) to EEG signals for detecting treatment resistance.

Purpose of the Study:

• To investigate the efficacy of a deep learning (DL) approach using GoogleNet convolutional neural network (CNN) on EEG data for detecting treatment resistance in depression.
• To identify distinctive EEG patterns associated with treatment resistance using Class Activation Maps (CAMs).

Main Methods:

• A deep learning model (GoogleNet CNN) was applied to EEG data from 77 patients with TRD, 43 with non-TRD, and 40 healthy controls.
• Class Activation Maps (CAMs) were utilized to visualize and analyze discriminative regions in the EEG data for TRD classification.
• Model performance was evaluated through direct classification accuracy and external validation.

Main Results:

• The GoogleNet model achieved high classification accuracies: 88.43% (healthy vs. non-TRD), 89.73% (healthy vs. TRD), and 90.05% (TRD vs. non-TRD).
• External validation for TRD-non-TRD classification yielded 73.33% accuracy.
• CAM analysis indicated that the TRD group exhibited dominant features across most electrodes within the DL architecture.

Conclusions:

• EEG-based deep learning demonstrates significant potential for classifying treatment resistance in depression.
• This methodology could become a valuable tool in psychiatric practice for early identification of patients requiring more aggressive treatment strategies.