FIELD: medicine.
SUBSTANCE: invention relates to systems for support of medical decision-making; it can be used for obtainment of auxiliary information, in particular, detection of signs of diabetic retinopathy in patients with diabetes mellitus, determination of a stage of the disease, and assessment of the efficiency of treatment, as well as determination of the risk of diabetic macular edema. A method is proposed, in which, using deep convolutional neural networks, on a colorful photo of the fundus, taken using a fundus camera and preliminary processed, detection of the fundus, optical disc, and macule center is performed. Then, a section with the fundus is cut out of the initial image, its size is changed, and the resulting image is cut for rectangular sections by a grid of 8 into 8 sections. After that, parameters of contours of detected signs are determined, for example, location, area, and a number of signs. At the same time, by means of a system of fuzzy inference, at least 6 different grades of visual signs of diabetic retinopathy and diabetic macular edema and one grade of shooting defects are detected, which are used then for visualization displayed to a doctor and for assessment of different parameters of detected signs with subsequent assessment of probabilities of the presence of stages of diabetic retinopathy and the presence of solid exudates in the macule.
EFFECT: invention provides an increase in a capability of segmentation of at least 6 different grades of DR signs and provision of a doctor with more complete information obtained during the operation of a system, for example, the probability of the presence of each of stages of diabetic retinopathy, the presence of solid exudates in the macule, its size, and amount of hemorrhage in each of quarters of the fundus for increasing the quality of diagnosis by the doctor.
4 cl, 5 dwg
