Facial Image-based Depression Detection Using Transfer Learning: A Resnet-18 Approach

Abstract

Depression Is One Of The Most Prevalent Mental Health Disorders Worldwide, Often Going Undiagnosed Due To The Lack Of Accessible And Objective Screening Methods. Recent Advancements In Artificial Intelligence And Computer Vision Have Enabled The Development Of Automated Systems Capable Of Detecting Depressive Symptoms Through Facial Analysis. This Paper Presents A Review And Implementation Framework For Depression Detection Using Facial Images, Leveraging Transfer Learning With The ResNet-18 Architecture In MATLAB 2024b. Emotion-labeled Datasets Such As FER-2013 And FER+ Are Utilized, With Emotion Classes Mapped Into Binary Categories Of Depressed And Non-depressed. The Proposed Methodology Includes Image Preprocessing, Data Augmentation, And Fine-tuning Of Pretrained Convolutional Neural Networks For Binary Classification. Synthetic Evaluation Results, Generated Due To Ongoing Model Training, Indicate An Expected Accuracy Of 91 % And An AUC Of 0.96, Demonstrating The Feasibility Of The Approach. This Study Also Provides A Comparative Analysis Of Existing Models, Discusses Limitations Such As Dataset Bias And Proxy Labeling, And Outlines Future Research Directions Including Multimodal Integration, Real-world Dataset Acquisition, And Explainable AI Techniques For Clinical Applicability. The Findings Suggest That Image-based Depression Detection Could Be A Scalable, Non-invasive Screening Tool To Assist Early Diagnosis In Both Clinical And Remote Healthcare Settings.

Keywords