Computerized lesion detection in colposcopy cervix images based on statistical features using bayes classifier

Abstract

Colposcopy is one of the methods for cervical cancer screening that uses visual testing based on the color change of abnormal cells that turn white when exposed to acetic acid called AacetoWhite (AW) region. In this paper, a novel approach to detect the AW region in the cervix image based on statistical features and Bayes classifier is presented. Colposcopic images are acquired in raw form, contains cervix, regions outside the cervix and parts of the imaging devices. In the preprocessing stage the irrelevant information in the cervical images are removed based on Mathematical morphology and Gaussian Mixture Modeling and also specularities are removed based on HSI colour space. The detection of lesion (AW region) is achieved by extracting the statistical features such as mean, standard deviation and skewness and the features are used as an input to the Bayes classifier. Segmentation results are evaluated on 260 images of colposcopy. © 2012 Springer-Verlag GmbH Berlin Heidelberg.