Toward leukocyte recognition using morphometry, texture and color

Although blood cell morphological analysis has been progressively replaced by new technologies, it is often possible and necessary to diagnose some types of leukemia by visual inspection. In order to automate this process, pattern recognition methods are used in the present work for computer-aided diagnosis, considering a variety of morphometric features, expressing size and shape, color and texture, which are combined in order to achieve more accurate results. The large number of measurements implies selection methods to be applied to the identification of the most discriminative subset. This paper describes 62 cell morphological attributes extracted, selected and analyzed using a system under development. We validate our results by using the Weka (Waikato environment knowledge analysis) machine learning algorithms. The obtained results illustrate the importance of feature selection for improving the classification of leukocytes from blood smears.