Application of neural network information technology for recognition and classification of image presentations of renal cell carcinoma in chronic kidney disease to choose the optimal method of treatment

Abstract

The information technology of recognition and classification of imaging representations of RCC complicated CKD with use of a neural network is offered. Approaches to architecture design, teaching methods, data preparation for training, training and neural network testing are described. The structural-functional scheme of the neural network is developed, which consists of the input, hidden and output layer, each individual neuron is described by the corresponding activation function with the selected weights. The expediency of using the number of neurons, their type and architecture for the task of recognition and classification of image representations of oncological phenomena of the organism is shown. Data of patients with RCC of complicated CKD, research department of reconstructive and plastic oncourology of NIR, urological department of "Lviv regional hospital", urology department of Lviv urological regional medical - diagnostic center, were used as initial data, on the basis of real observations, a database for training and education of the neural network was formed. An analysis of the efficiency, speed and accuracy of the neural network, in particular, a computer simulation using modern software and mathematical modeling of computational processes in the middle of the neural network. Software has been developed for preliminary preparation and processing of input data, further training and education of the neural network and directly the process of recognition and classification. According to the obtained results, the developed model and structure of the neural network, its software tools show high efficiency both at the stage of preliminary data processing and in general at the stage of classification and selection of target areas of diseases. The next stage of research is the development and integration of software and hardware system based on parallel and partially parallel computer technology, which will significantly speed up computational operations, achieve the learning and training of the neural network in real time and without loss of accuracy. The presented scientific and practical results have a high potential for integration into existing information and analytical systems, medical analysis the choice of tactics for the treatment of patients with RCC complicated CKD, and health monitoring systems in the preoperative and postoperative periods.