Evaluation of the effectiveness of a medical decision support system based on optical coherence tomography data

Diseases of the retina, including age-related macular degeneration and diabetic macular edema, are among the leading causes of vision loss and blindness in older patient groups, while up to 80% of cases of vision loss can be prevented through early diagnosis and timely treatment. Interpretation of OCT images, which are the “gold standard” for the diagnosis of retinal diseases, requires significant physician expertise and is accompanied by interobserver variability. According to the literature, artificial intelligence technologies in ophthalmology make it possible to automate the processes of diagnosis and prediction of treatment outcomes based on OCT data, to create decision-making algorithms, and to obtain accurate and reproducible results in the tasks of detecting and measuring the quantitative characteristics of biomarkers. The aim of this study was to evaluate the effectiveness of applying a clinical decision support system based on optical coherence tomography data in the clinical practice of an ophthalmologist. The material of the study consisted of anonymized OCT images of patients with age-related macular degeneration and diabetic macular edema, as well as images with a normal retina. The decision support system includes classification models, segmentation models, and an algorithm for calculating quantitative parameters. The evaluation of effectiveness was carried out with respect to three types of annotation: disease diagnosis, identified biomarkers, and their quantitative measurements. The developed clinical decision support system based on OCT data achieves high accuracy in the classification of retinal diseases (Accuracy = 0.87, Precision = 0.92, Recall = 0.86) and in the segmentation of biomarkers (Dice coefficient = 0.84 for PED, 0.77 for SRF, and 0.72 for IRF). In addition, the system allows the time required for OCT image analysis to be reduced by 16-fold and 50-fold for an expert and a novice physician, respectively. Thus, the developed clinical decision support system based on OCT data provides high accuracy in the classification and segmentation of retinal diseases, corresponds to the level of reproducibility of expert annotation, and meets the requirements of the GOST standard. The implementation of the system in practice may increase the availability of ophthalmological care, improve the effectiveness of screening programs, and create a foundation for the application of telemedicine technologies in ophthalmology.