A.V. Samorodov

Morphometric research of cytological preparations (CP) is widely used in the laboratory analysis practice. The results are highly informative and allow evaluating conditions of a person or an animal organism at the cellular level. Results of cells analysis in cytological preparations reflects the functional state of cells and of human and are used to solve grate variety of tasks: from screening analysis and early diagnostics to differential diagnostics in oncology, haematology and other fields of medicine. However until recently the unified approach to the evaluation of the structure and form of images has not been elaborated, requirements on accuracy of this analysis have not been defined, and there are no corresponding standards. Meanwhile to obtain statistically reliable results it is necessary to analyze the significant amount of information contained in the CP image that complicates the visual analysis. Consequently, it is impossible to widely introduce CP morphometric research in different fields of science. To overcome the specified shortcomings it is necessary to develop a method of CP quantitative morphometry that will provide for its automation. The automation of CP morphometric analysis is based on measuring not only quantitative parameters of objects (sizes), but also their qualitative properties (form, structure). The tasks and steps of automated CP images analysis include: CP quality evaluation, integral morphometric parameters estimation, cells detection, cells segmentation, cells classification, individual cells morphological parameters estimation, and fusion of all the results for diagnistical decision. The research carried out in BMSTU has shown that the method of spatial-frequency analysis allows to formalize the description of the structure and form of MBP images. Criteria of the quality of CP preparing have been provided, the range of characteristics of cells morphology is worked out, the classification method of cells by means of probabilistic features is developed. One of the less considered tasks in literature is systematic investigation of cells detection methods. In presented work different approaches to cells detection are considered. All detection methods were joined into three groups: template-based, non-template-based, and hybrid. Real-time algorithm for cells detection, based on hybrid approach was worked out and tested in blood smears analysis for leukocyte detection. The results shows, that detection rate is 95% when tested on blood smears, prepared by different laboratory assistants. The work presented was carried out by support of Russian President Grant MK-6323.2008.8.