Investigation of Textural Features for the Problems of Bone Marrow Cell Recognition in Information-Measuring Systems of Oncohematology

The influence of the parameters of constructing a matrix of spatial adjacency on textural features in the problems of recognizing bone marrow cells in information-measuring systems for diagnosing acute leukemia has been studied. 100 images of blast cells of B- and T-cell acute lymphoblastic leukemias were examined. Five textural features calculated on the basis of the spatial adjacency matrix are considered – energy, moment of inertia, local uniformity, maximum probability, entropy. When constructing an adjacency matrix, the variable parameters were analyzed – the type of the color component of the RGB-model of the color image, the distance and direction of the adjacency. For a given sample of images, the range of adjacency distances in which the largest change in the values of texture features is observed was 1–11 pixels. The values of features of different types vary in the range of 20–1700%. The maximum information content was obtained for the green component of the color image for the texture feature “local uniformity” (information content coefficient of 0.48) with an adjacency distance of 1 pixel. For practical use, it is recommended to use four directions of adjacency when constructing spatial adjacency matrices. The results obtained are important for specialists working in the field of designing information and measuring systems for hematology oncology (diagnostics of dangerous oncological diseases – acute leukemia).