Selection of image transformations in the computer analysis of cytological specimens

The paper considers the problem of the diagnostic analysis of blood system tumors (hemoblastosis) using special techniques. Images of lymph node specimen are taken from three groups of patients with such diagnoses as indolent chronic lymphatic leukemia, transformed chronic lymphatic leukemia, and de novo large and mixed cell lymphomas serving as source data. The analysis of the feature description showed that different sets of informative features correspond to different diagnoses. Thus, special techniques of image analysis and recognition, which allow taking into account the informative nature of images when selecting image transformation algorithms are required. Such possibilities are offered by the technique of image transformation selection when solving recognition problems. This technique allows taking into account the description peculiarities of each class of images and utilizing the recognition algorithm appropriate for the given class. The paper shows that application of this technique helps substantially improve the accuracy of the recognition.