Solving classification problems for ordered alternatives on the example of application in ophthalmology

Abstract

The aim of this study is to develop a solution to classification problems for ordered alternatives in the field of ophthalmology by the example of determining the stages of development of primary open-angle glaucoma. Materials and methods: the development of the method was carried out based on a detailed medical history of patients who were under observation at the Center for Laser Vision Restoration “Optimed”, Ufa. The database represents 793 unique patients. The main criterion for inclusion in the analysis was the diagnosis of “Primary open-angle glaucoma” (POAG) (H40.1) with the appropriate stage of the disease. The model was trained by the method of ordered regressions, in which the principle of reverse selection was used:__ all variables were included in the regression equation, and then statistically insignificant variables were excluded in the reverse order. The selection of the best model was made based on the minimum values of the information criteria Akaike, Schwartz. When analyzing the accuracy of the prediction, the conjugacy matrix was analyzed and the accuracy metric for each class was calculated. The overall accuracy metric of the classifier was defined as the weighted average of the volumes of each class represented in the test sample. Also, based on the ophthalmologists’ opinion, errors for each class were weighted by 0.9. Results: significantly associated with the stage of primary open-angle glaucoma disease were identified. The model best predicts the second stage of glaucoma. This can be justified by the size of the sample, since more than 40 percent of the sample included patients who were diagnosed with second-degree glaucoma. The overall weighted accuracy of the glaucoma stage classifier is 0.602. Conclusion: the developed solution allows you to speed up and improve the procedure for determining the stage of glaucoma but requires further improvement.