Moskalenko A. Models and methods of information technology for radionuclide diagnostics of pathologies under incomplete definiteness

Research object - weakly-formalized process of classification of results of radionuclide examination of the kidneys and myocardium under conditions of incomplete certainty, caused by arbitrary initial conditions of images obtain; goal of research - increase a functional efficiency of radionuclide diagnosis system of renal and myocardial pathology of results of testing with gamma-camera by creation an information technology of machine learning; methods of research - methods of system analysis and information technology intellectual data analysis, probability theory and mathematical statistics, information theory, object-oriented methodology of design of dataware and software; the results - the important scientific and practical task of improving the functional efficiency of information system of radionuclide diagnostics of renal and myocardium abnormalities is solved; novelty - the model and machine learning method, which based on the using of multi-interval control tolerances on the value of quantitative features and frequency of categorical diagnostic features, and hierarchical container of classes are first developed, which, unlike existing account for multimodality of the probability distribution of categorical and numerical diagnostic data, that it is allowed to increase recognition accuracy of myocardium and kidney functional states; the first time it is proposed an algorithm for evaluation of information criterion of optimization parameters of the diagnostic system, which carries out training system radionuclide diagnostics with a hierarchical structure of containers of classes, thus improving the reliability of decision rules; enhanced models and methods for ensemble cluster analysis using the information criterion for optimizing individual and the resulting partitions, that make possible to increase the stability of the results of cluster analysis and the functional efficiency of machine learning; the swarm search method of global maximum of functional efficiency information criterion by modifying the procedures for updating personal and global best particle swarm positions which allow improves the efficiency of finding the optimal in an information sense set of features with minimal power gained further development; the degree of implementation - Institute for Scintillation Materials NAS of Ukraine (Kharkiv), Sumy Regional Clinical TB Dispensary, learning process of Computer Sciences of Sumy State University; field of application - information technology.