Glaucoma Detection for Automated Referral System: Using OCT Data and Fine-tuning LLM Models

Glaucoma is a leading cause of irreversible blindness, requiring early detection and timely referrals. This study proposes an AI-powered automated system utilizing the GPT-4o and GPT-4o-mini as large language models (LLMs) to generate referral letters for glaucoma-suspicious cases based on clinical data (220 OCT images). The system extracts key patient information, interprets ocular examination results, and formulates structured referral letters with standardized clinical terminology. By reducing administrative workload and improving communication between general ophthalmologists and glaucoma specialists, this approach enhances efficiency in glaucoma management. The proposed system aims to streamline referral pathways, ensuring timely specialist intervention and improving patient outcomes in early glaucoma detection and care. The proposed system achieved a 100% recall (sensitivity), successfully identifying all glaucoma cases. Among the models evaluated, GPT-4o demonstrated promising performance, achieving an accuracy of 91%, precision of 88%, and an F1score of 93%, outperforming GPT-4o-mini.Clinical Relevance- Early glaucoma detection relies on timely and accurate referrals to a specialist in its early stages to prevent irreversible vision loss. The proposed AI-powered system utilizes a LLM using a vision API to automate the generation of structured referral letters based on OCT data, reducing the administrative burden for healthcare providers. This approach streamlines referral pathways directly from the OCT device, enabling operators to facilitate early intervention and ultimately improve patient outcomes. Integrating AI into glaucoma care has the potential to enhance diagnostic accuracy, optimize resource utilization, and support more efficient clinical decision-making.