A texture approach to leukocyte recognition

Despite its ubiquity in image data, a formal approach or precise definition of texture does not exist [1]. An acceptable approach is to describe texture as a repeating pattern of local variations in image intensity which are too fine to be distinguished as separate objects at the observed resolution [2]. Texture has been widely used in image segmentation, shape analysis, estimation of orientation and other inspection tasks as classification of cell micrographies [3,4]. Textural features can be statistically evaluated using Grey Level Cooccurrence Matrices (GLCM)[1], a joint probability distribution of two picture elements in a given direction, distance and block size (window). Many measurements can be extracted from the isotropic GLCM addressing contrast (amount of local variations) and orderliness (regularity of the pixel values within the window). Humans often use qualitative descriptions of the texture, which occurs in leukemia diagnosis frequently. It has been observed in the laboratory of Hematology of Ribeirão Preto the usefulness of applying quantitative analysis on medical images, once an automatic pattern recognition classifier allows faster objective quantifications at lower costs. The automation of this process would permit competent people to devote their attention to more demanding tasks while pattern recognition tests would be performed in the laboratory routines. Another potential application of a software in leukocyte recognition is for educational purposes. This paper presents an application of Haralick’s techniques [1] in order to enhance the characterization of blood smear images through texture information addition. We start by presenting a pattern recognition software under development [5], its limitations in terms of cytoplasm segmentation using color criteria and how textural data can improve the system accuracy. Henceforth, parameters are proposed for tuning the function of the angular relationship between the neighboring resolution cells as well as the distance between them. Texture-based parameters extracted from GLCM are entropy, energy, inertia and inverse different moment and these are interpreted in terms of its mean and standard deviation for leukocyte differention. Promising results are presented, including feature space distributions using some of the more relevant considered measures.