Semi-automated lession grading in cervix images with Specular Reflection removal

Colposcopy is the method for cervical cancer detection at early stage which is examined by colposcopist based on the changes in the color of cervix tissues. When gynecological exam is done, over the uterine cervix surface 3 % acetic acid is applied for two minutes, the method is known as visual inspection with the application of acetic acid (VIA), which causes whitening of precancerous tissues known as malignant regions of the epithelium; these regions of tissue are called acetowhite lesions (AW). During the treatment of cervical cancer colposcopic image is obtained with a specialized camera outfitted with a colposcope with green filters. Due to illumination of light on cervix tissue image preprocessing is required prior to applying AW lessions detection algorithms on colposcopic images to remove Specular Reflections (SR) and to differentiate the cervix region-of-interest (ROI) from other image regions which are not relevant to the analysis. In this work by preprocessing SRs are identified and then SRs are removed. After preprocessing to identify the malignant regions a texture segmentation is used by applying k-means clustering for automatic separation of non-cancerous and precancerous regions of cervix tissue. This algorithm would help the gynecologist during the treatment of cervical cancer.