Author/Editor		Žganec, Mario
Title		Postopki za razvrščanje vzorcev in prikaz rezultatov pri analizi celičnih jeder s slikovnim citometrom
Type		monografija
Place		Ljubljana
Publisher		Medicinska fakulteta
Publication year		2003
Volume		str. 146
Language		slo
Abstract		Image cytometry uses sophisticated measurements of chromatin distribution in a cell nucleus to gather information about physiological and pathological processes in cells, mostly related to detection of cancer. Automatic recognition and classification of cells, based on their images, is an essential part of image cytometry. Since large amounts of data have to be manipulated and complex computations performed in the process of data analysis, special care must be taken to facilitate these manipulations and to present data in an appropriate way. Only a well-designed user interface can allow for fusion of the knowledge and experience of an expert with exact mathematical computations and measurements performed by the machine. The aim of this thesis is to develop and test pattern recognition algorithms, that are appropriate for classification of cell samples, and to build a user interface that will improve the interaction between the user and the computer, mainly by presenting the data in a comprehensive way. We concentrated on the pattern recognition algorithms, based on statistical decision ,procedures with gaussian models of class probability distributions. Special care was taken to optimize the learning algorithms for speed. Fast learning of the recognizer allows for better and easier feature selection, by simply testing all the possible feature combinations. New features derived by ortogonalization of the feature covariance matrix were tested as well. The recognition rate when using these new features was similar to the one obtained with the original feature set. Data manipulation, image recognition and data presentation algorithms were designed to maintain algorithm execution times within acceptable limits even when analyzing thousands of slides and millions of cells. The user interface was designed to be simple and intuitive yet providing the user with complete and extensive information. (Abstract truncated at 2000 characters).
Descriptors		NEOPLASMS IMAGE CYTOMETRY CELL NUCLEUS PATTERN RECOGNITION MATHEMATICAL COMPUTING DECISION SUPPORT TECHNIQUES NORMAL DISTRIBUTION ROC CURVE BREAST NEOPLASMS LUNG NEOPLASMS