An Interpretable and Expandable Deep Learning Diagnostic System for Multiple Ocular Diseases: Qualitative Study

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Current AI in medicine lacks transparency in diagnostic reasoning.
• Automatic disease diagnosis platforms often fail to explain their decision-making processes.

Purpose of the Study:

• To create an interpretable and expandable AI framework for diagnosing multiple ocular diseases.
• To provide tailored treatment recommendations for individual patients.

Main Methods:

• Utilized annotated ophthalmic images, decomposed by anatomical knowledge.
• Developed a deep learning framework with four diagnostic stages: identification, localization, classification, and treatment advice.
• Integrated a telemedical system for clear communication of diagnostic reasoning.

Main Results:

• Achieved high accuracy in disease identification (93%) and anatomical localization (up to 90%).
• Demonstrated strong performance in classifying specific conditions (79%-98%) and recommending pterygium treatment (>95%).
• Successfully built an interpretable AI platform that clarifies the diagnostic workflow for clinicians and patients.

Conclusions:

• The developed AI platform enhances diagnostic transparency and provides actionable treatment suggestions.
• The system is expandable for new diseases and aids in clinical training for junior doctors.
• This framework can improve access to quality eye care in resource-limited and remote areas.