Three-Stream Convolutional Neural Network for Depression Detection With Ocular Imaging.

Area of Science:

• Ophthalmology and Neuroscience
• Artificial Intelligence in Healthcare
• Computational Psychiatry

Background:

• Depression is a widespread mental disorder with significant impacts on mental and physical health.
• Current depression diagnosis relies on subjective clinical assessments, risking misdiagnosis due to a lack of objective biomarkers.
• Eye movement patterns and pupil dilation are emerging as potential objective indicators of emotional and cognitive dysfunction in depression.

Purpose of the Study:

• To develop an automated method for depression detection using ocular imaging data.
• To explore the utility of spatio-temporal and semantic features from eye movement for depression diagnosis.
• To overcome limitations of previous studies relying on manually extracted eye movement features.

Main Methods:

• Proposed a novel Three-Stream Convolutional Neural Network (TSCNN) model for depression detection.
• Leveraged raw ocular imaging data, incorporating spatio-temporal information via optical flow with varying sampling intervals.
• Integrated semantic information from paradigm images using an encoder as prior knowledge for the classification task.

Main Results:

• The TSCNN model achieved a classification accuracy of 79.3% on a self-collected dataset.
• The method effectively utilizes both dynamic (spatio-temporal) and static (semantic) features from ocular imaging.
• Demonstrated the potential of deep learning on raw ocular data for objective depression assessment.

Conclusions:

• The proposed TSCNN method offers a promising, data-driven approach to depression detection.
• Ocular imaging, analyzed with advanced AI, can serve as a valuable tool in diagnosing depression.
• This research highlights significant potential for future clinical utility in objective mental health diagnostics.