Martyanova T. Method and System for Determining the Fundus Vascular Lesions in Diabetes

The object of the research is the process of recognition of the fundus abnormalities in diabetes with computer system. The aim is to improve the reliability of analysis biomedical data by creating a diagnostic method based on the quantify the geometry of objects and system of diagnostic a fundus abnormalities in diabetes. Methods: methods of digital processing of signals and images to their previous treatment, the theory of computer systems and networks in modeling of processes of clustering and pattern recognition methods for programming languages of high and low levels when programming graphics cards GPU, morphological segmentation in the development of mathematical models, mathematical and computer modeling for experimental studies. Theoretical results: improved method for processing of biomedical images based on the spatial spectrum of connectivity using mathematical tools morphological segmentation and introduction hidden corner color, which reduced susceptibility to deformation, which included the formation of an image and its noise. Firstly, there had been developed the mathematical model for calculating the area of pathology using geometric shapes of optimal choice for circuit object. Due to the introduction of rationing criterion that determined experimentally and must be equal to or greater than the value of 0.85, there had been selected algorithm of optimization the geometrical shapes that define the zone area of pathology. Got the further development the optimal clustering method of fundus pathologies through the use of normalization, which allows for non-typical set of all factors of equal weight, allowing experimentally determine the number of clusters 8, and stop the clustering option 0.5. Got further development the system of image clustering fundus, which use the phase correlation between images and allow more reliably recognize images of the fundus, namely 23.8%. Practical results are: development of algorithmic and software and hardware to detect pathologies of biomedical images of the fundus, which is based on a range of spatial coherence fundus with the aid of morphological segmentation and introduction hidden corner color, to effectively recognize and cluster images. Based on the results of theoretical and practical studies there had been formed the methodology of experimental studies fundus pathologies. Based on the developed technique of experimental studies there had been chosen hardware architecture developed system for recognition. The degree of implementation - the results of the thesis had been introduced in the Vinnitsa Regional State Administration and the Vinnitsa Oblast Children's Clinical Hospital. Scope (industry) use - medicine.