TCEDN: A Lightweight Time-Context Enhanced Depression Detection Network

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automatic video recognition of depression is crucial for clinical applications.
• Traditional models face challenges: high computational cost, poor facial movement feature effectiveness, and spatial feature degradation.

Purpose of the Study:

• To propose a lightweight Time-Context Enhanced Depression Detection Network (TCEDN) to address limitations of existing depression recognition models.
• To enhance the precision and reduce computational complexity of video-based depression detection.

Main Methods:

• Utilized attention-weighted blocks to aggregate and enhance video frame-level features, reducing computational load.
• Integrated temporal and spatial changes of raw video and facial movement features using self-learning weights for improved precision.
• Employed a fusion network combining 3-Dimensional Convolutional Neural Network (3D-CNN) and Convolutional Long Short-Term Memory Network (ConvLSTM) to minimize spatial feature loss.

Main Results:

• Achieved performance on par with state-of-the-art techniques on AVEC2013 and AVEC2014 datasets for depression detection.
• Demonstrated significantly lower computational complexity compared to mainstream methods.

Conclusions:

• The proposed TCEDN effectively detects depression from videos with high accuracy and efficiency.
• The network offers a promising, computationally efficient solution for clinical depression recognition applications.