DPD (DePression Detection) Net: a deep neural network for multimodal depression detection.

Area of Science:

• Artificial Intelligence
• Computational Psychiatry
• Machine Learning for Healthcare

Background:

• Depression diagnosis is complex and subjective, often relying on physician interviews.
• Existing automatic depression detection models lack robustness due to single-dataset testing.

Purpose of the Study:

• To propose deep learning models for automatic depression detection using multimodal data.
• To enhance the robustness and scalability of depression detection systems.
• To assist in the clinical diagnosis of depression.

Main Methods:

• Developed DePressionDetect Net (DPD Net), a Graph Neural Network-enhanced Transformer model, integrating textual, audio, and visual features.
• Proposed DePressionDetect-with-EEG Net (DPD-E Net) incorporating Electroencephalography (EEG) signals and speech data.
• Evaluated models in clinical and social media settings across four benchmark datasets.

Main Results:

• DPD Net and DPD-E Net outperformed state-of-the-art models on three datasets (E-DAIC, Twitter, MODMA).
• Achieved competitive performance on the D-vlog dataset.
• Ablation studies confirmed the effectiveness of proposed modules and multimodal integration.

Conclusions:

• The proposed deep learning models offer a more robust and scalable approach to automatic depression detection.
• Combining diverse data modalities significantly enhances depression detection accuracy.
• These models can serve as valuable tools to assist in depression diagnosis.