Discovering Important Regions of Cytological Slides Using Classification Tree

Modern digital microscopy systems allow imaging of biological material with very high accuracy. Paradoxically, this gives rise to many problems because huge amounts of raw data significantly increase the time required by specialist to analyze them. As a result, we obtain a time-consuming diagnostic process and reduction of the number of patients being diagnosed. The paper presents a method of discovering regions of the cytological image, which are essential to correct diagnosis. The purpose of this method is to help pathologists by indicating regions of the image that should be analyzed first. Moreover, method can be used to explore new, previously unknown features discriminating benign from malignant lesions. Multi-level image thresholding is responsible for image segmentation and is the core of the proposed system. Thresholds are evaluated by predictive accuracy on testing dataset. Honey Bee Mating Optimization (HBMO) algorithm is applied to find the optimal threshold set. The developed method was successfully applied to analyze cytological images of biological material collected from a breast tumor.