COMPUTER-AIDED DIAGNOSIS OF BREAST CANCER USING GAUSSIAN MIXTURE CYTOLOGICAL IMAGE SEGMENTATION 2
Marek Kowal, Paweł Filipczuk, Andrzej Obuchowicz, Józef Korbicz · 2011
This paper presents an automatic computer system to breast cancer diagnosis. System was designed to distinguish benign from malignant tumors based on fine needle biopsy microscope images. Studies conducted focus on two different problems, the first concern the extra ction of morphometric and colorimetric parameters o f nuclei from cytological images and the other concentrate on bre ast cancer classification. In order to extract the nuclei features, segmentation procedure that integrates results of a daptive thresholding and Gaussian mixture clusterin g was implemented. Next, tumors were classified using four different classification methods: k-nearest neigh bors, naive Bayes, decision trees and classifiers ensemble. Dia gnostic accuracy obtained for conducted experiments varies according to different classification methods and f luctuates up to 98% for quasi optimal subset of fea tures. All computational experiments were carried out using microscope images collected from 25 benign and 25 malignant lesions cases. According to the National Cancer Registry in Poland breast cancer is the most common cancer among women. In 2008, there were 14,576 diagnosed cases of breast cancer in Polish women. Out of these cases, 5362 deaths were the result. There has also been an increase of breast cancer by 3-4% a y ear since the 1980's. The effectiveness of treatment la rgely depends on early detection of the cancer. Important and often used diagnostic method is so-ca lled triple-test. It is based on three medical examinations and allows to achieve high confidence of diagnosis. The triple-test includes self examination (palpation), mammography or ultrasonography imaging and FNB (Fine Needle Biopsy). FNB is collecting nucleus material directly from tu mor. Obtained material is examined under a microscope to determine the prevalence of cancer cells [29]. The present approach requires a deep knowledge and experience of the cytologist responsi ble for diagnosis. Automatic morphometric diagnosis can make the decision objective and assist inexperi enced specialist. It can also allow screening on a large scale where only difficult and uncertain cases woul d require additional human diagnosis. Along with th e development of advanced vision systems and computer science, quantitative cytopathology has become a useful method for the detection of diseases, infect ions as well as many other disorders [9, 28]. In th e literature one can find approaches to breast cancer classification [5,6,10,13,14,16,17,20,21,24,26]. Mentioned approaches are concentrated on classifyin g FNA (Fine Needle Aspiration) or FNB slides as benign or malignant. In this paper, we present a decision support system that allows distinguish malignant from the benign breast tumors. The classification of the tum or is based on morphometric examination of cell nuc lei [29,30]. In contrast to normal and benign nuclei, w hich are typically uniform in appearance, cancerous nuclei are characterized by irregular morphology th at is reflected in several parameters described in detail further in the article. Features were extracted fro m segmented images obtained by hybrid segmentation method based on Gaussian mixture clustering and adaptive thresholding. The quality of segmentation and feature extraction was tested by using the set of classifying algorithms. The measure is based on classification accuracy obtained by leave-on-out cross-validation. In this work four different classification methods were used to rate the feature subsets: k-nearest neighbors, naive Bayes classifier, decision trees a nd classifiers ensemble. The paper is divided into four sections. Section 1 gives an overview of breast cancer diagnosis techniques. Section 2 describes the process of acqu isition of images used to breast cancer diagnosis.