An Explainable AI Framework for Corneal Imaging Interpretation and Refractive Surgery Decision Support.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Corneal topography analysis is crucial for diagnosing ectatic diseases like keratoconus and planning refractive surgery.
• Current interpretation methods can be complex and time-consuming, necessitating advanced decision support tools.

Purpose of the Study:

• To develop and validate an explainable neuro-symbolic and large language model (LLM)-driven framework for corneal topography interpretation.
• To provide intelligent decision support for early keratoconus detection and refractive surgery planning.

Main Methods:

• A four-stage pipeline involving automated parameter extraction, knowledge graph mapping, Bayesian inference, and LLM-based reporting (DeepSeek, GPT-4.0).
• Prospective analysis of 20 eyes using IOLMaster 700 data.
• Benchmarking against senior corneal specialists and baseline Convolutional Neural Network (CNN) and Vision Transformer (ViT) models.

Main Results:

• The framework achieved high diagnostic performance for early keratoconus detection (92-94% sensitivity/specificity, AUC 0.95).
• It demonstrated strong performance for refractive surgery eligibility (F1 score 0.90).
• Generated bilingual reports were highly rated for clarity and clinical utility, with outputs generated rapidly (~95 seconds).

Conclusions:

• The neuro-symbolic and LLM-driven framework offers a transparent, rapid, and clinically actionable AI solution for ophthalmic practice.
• This approach significantly enhances early ectatic disease detection and supports individualized refractive surgery planning.
• The integration of explainable AI (XAI) principles ensures trustworthy and interpretable clinical decision support.