Segmentation of blood vessels in colposcopic images using polarized light and Sauvola thresholding

Abstract

Cervical cancer is one of the two most common gynecological cancers in the world, including breast cancer. Signs of cervical disease are usually the presence of atypical epithelium, superficial bleeding or abnormal vascular proliferation. Most of these signs are directly related to cervical intraepithelial neoplasia (CIN) and cervical cancer. Currently, to detect epithelial lesions as well as to observe the shape of blood vessels, the main diagnostic methods used are colposcopy and visual examination. This method has low sensitivity and specificity because subjective factors still exist and the method does not clearly distinguish the shape of proliferating blood vessels. Therefore, in order to improve the efficiency of disease diagnosis, many studies applying image processing techniques to support auto-diagnosis have become topics of interest. However, studies that support automatic identify abnormal blood vessel shape and density are very limited. In this study, colposcopy images were recorded by digital colposcopes. These images are taken under polarized light to help reduce reflections from the surface and support for better image processing steps. Then, Sauvola threshold method is used to separate blood vessels on the surface of the cervix. It is combined with three different image preprocessing methods to enhance the contrast between the blood and the background. Finally, the sensitivity and specificity of these methods were calculated and evaluated. The results of the study set the stage for cervical blood vessel identification studies as well as cervical cancer assessment.