Code-Free Machine Learning for the Detection of Common Ophthalmic Diseases

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR), age-related macular degeneration (AMD), and glaucoma are leading causes of vision loss.
• Early detection and diagnosis are crucial for effective treatment and prevention of irreversible vision impairment.
• Developing accessible tools for disease detection can empower clinicians and improve patient care.

Purpose of the Study:

• To evaluate a code-free machine learning (ML) method for detecting DR, AMD, and glaucoma from fundus photographs.
• To enable physicians without programming experience to develop diagnostic ML models.
• To assess the performance of ML models in identifying specific retinal pathologies.

Main Methods:

• Two classification models (binary and multi-class) were developed using Google Vertex AI's no-code AutoML Vision platform.
• A dataset of 800 fundus images was used for development and internal validation.
• External validation was performed on 400 images from a separate dataset, evaluating performance using AUPRC, precision, recall, and accuracy.

Main Results:

• The binary model achieved an AUPRC of 0.967 internally and 92.3% accuracy externally.
• The multi-class model achieved an AUPRC of 0.906 internally and 90% overall accuracy externally.
• Both models demonstrated high precision and recall, indicating reliable performance in disease detection.

Conclusions:

• Code-free ML approaches empower physicians to create diagnostic tools for retinal diseases without coding expertise.
• This facilitates earlier detection and intervention, potentially reducing vision loss.
• The study highlights the potential of accessible AI tools in clinical practice and bridging the gap between research and deployment.