MANUAL PROCESSING OF MEDICAL IMAGES BY A DIAGNOSTICIAN IN A TOMOGRAPHIC AND X-RAY RECOGNITION SYSTEM FOR THE SEARCH AND LOCALIZATION OF PATHOLOGIES

Abstract

localization of pathologies. In this research paper, it is necessary to develop a system of recognition of tomographic and X-ray images for finding and localization of pathologies and to consider the principles of manual processing of medical images by a diagnostician. To find and localize anomalies on tomographic and X-ray images, a system is proposed which will consist of the following units: the patient information entry unit; a medical images processing unit comprising: a convolutional neural network (CNN) for the classification of tomographic and X-ray images; CNN to determine the dependence of the values of the spatial factor Wz on the standardized z-values with further calculation of the coefficient of the total spatial autocorrelation; CNN for segmentation of tomographic and X-ray images; the manual image processing subsystem presented by the diagnostician;

An opinion unit, which includes a neural network (NN) designed to compare the results obtained; a unit for classification of detected pathologies, including NN; database of both existing images and new ones, including with the result of processing; report preparation unit. This system includes the following units: patient information entry, medical images processing for delivering an opinion, for classification of pathologies identified, database, report preparation. In the proposed system, the initial step is to obtain tomographic or X-ray images, which are then fed to the patient information entry and medical images processing units. As a result of data entry, patient information is included in the database along with tomographic or X-ray images. In the proposed system of recognition of tomographic and X-ray images, for finding and localization of pathologies it is possible for the diagnostician to highlight the area suspicious from his/her point of view and further process only this area or, using existing convolutional neural networks, identify areas of pathologies - neoplasms, or choose particular algorithms for medical images processing.