Developing an approach to differential diagnosis of eye diseases based on knowledge formalization

Objective: to develop and explore an approach to differential diagnosis of eye diseases based on knowledge formalization, aimed at application in primary health care. Materials and methods. We proposed a formal model of the subject area, including sets of symptoms, diagnostic hypotheses, linguistic variables, membership functions, inference rules, and a priori assessments of the disease. We developed a hybrid method of diagnostic inference, combining fuzzy logic and Bayesian probability hypothesis refinement. This method served as a basis for the creation of a decision-making system. An experimental evaluation was conducted at the Department of Clinical Diagnosis of the Eye Disease Clinic (Razumovsky State Medical University of Saratov). The results of patient examinations performed by ophthalmologists with at least three years of experience as part of routine practice were compared with the decision-making system conclusion. Results. The proposed approach ensured the formation of a ranked list of diagnostic hypotheses and an interpretable diagnostic conclusion. In noise-free scenarios, the mean diagnostic accuracy was 82.86%, while in the presence of noise and with contradictory symptoms, it amounted to 85.71% and 76.19%, respectively. This approach demonstrates its superior robustness and lower computational cost vs. a solution based on a large language model. Conclusion. The developed approach allows for the formalization of diagnostic knowledge, accounts for the ambiguity and incompleteness of clinical information, and ensures the interpretability of diagnostic inference. The obtained results confirm the potential of this approach for use in primary care decision support systems.