Semantic Graph Interpretation of Corneal Topography Using Abstract Meaning Representation for Surgical Planning in Pterygium-Associated Cataract

Pterygium-induced corneal distortion poses a significant challenge in cataract surgery, as localized changes in corneal curvature can lead to astigmatism and compromise intraocular lens (IOL) power accuracy. Preoperative corneal topography and biometry are therefore essential, particularly when astigmatism exceeds 1.5 diopters (D), which often necessitates pterygium removal before cataract surgery. However, current interpretation of topography maps remains largely manual and subjective. This study introduces an Abstract Meaning Representation (AMR)-based framework for transparent, machine-readable interpretation of corneal topography. Ophthalmologists’ textual descriptions, such as “paranasal pterygium causes 2.0D with-the-irregular astigmatism”, were converted into semantic graphs that encode causal, spatial, and quantitative relationships. Each AMR node was aligned with corresponding topography parameters (K1, K2, Km, ΔK) using graph neural networks and transformer-based AMR embeddings to integrate linguistic and imaging data for explainable reasoning. The resulting framework accurately captured the relationship between pterygium features and induced astigmatism, supporting AI-assisted recommendations for preoperative management. Compared with conventional CNN or ViT-based models, the AMR system demonstrated superior interpretability, allowing clinicians to visualize the reasoning behind surgical decisions. Integrated with ocular surface video analytics and the meHealth (mobile eye health) platform, this approach enables multimodal, context-aware decision support and lays the foundation for combining AMR with large language models (LLMs) for automated report generation and clinical dialogue systems, advancing the development of transparent, patient-centered ophthalmic artificial intelligence.