Explainable Artificial Intelligence Framework for Predicting Treatment Outcomes in Age-Related Macular Degeneration.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Informatics

Background:

• Age-related macular degeneration (AMD) is a primary cause of irreversible vision loss.
• Current methods for predicting AMD treatment outcomes lack transparency and flexibility.
• There is a need for advanced decision support systems in AMD management.

Purpose of the Study:

• To develop and validate a hybrid neuro-symbolic and large language model (LLM) framework for explainable AMD treatment prognosis.
• To integrate mechanistic disease knowledge with multimodal ophthalmic data for improved predictive accuracy.
• To provide transparent, evidence-based risk stratification for personalized AMD treatment planning.

Main Methods:

• A pilot cohort of ten AMD patients undergoing surgical management was studied.
• Multimodal data including clinical documents and ophthalmic imaging (OCT, FA, SLO, B-scan US) were collected and processed.
• A neuro-symbolic approach combined a domain-specific ophthalmic knowledge graph with an LLM fine-tuned on ophthalmology literature and EHRs.

Main Results:

• The hybrid model achieved high performance on an independent test set (AUROC 0.94, AUPRC 0.92, Brier score 0.07), outperforming baseline models.
• Explainability metrics demonstrated that >85% of predictions were supported by knowledge-graph rules and >90% of narratives cited key biomarkers.
• A case study showed accurate real-time risk stratification for treatment needs and potential complications.

Conclusions:

• The proposed framework effectively integrates multimodal evidence for explainable AMD treatment prognosis.
• This approach offers transparent causal reasoning and supports personalized treatment planning.
• The study establishes a scalable, privacy-aware template for next-generation decision support in AMD and other retinal diseases.