Identification of Abnormal Cervical Regions from Colposcopy Image Sequences

Cervical cancer is the third most common cancer in women worldwide and the leading cause of cancer death in women of the developing countries. Cancer death rate can be greatly reduced by regular screening. One of the steps during a screening program is the detection of the abnormal cells that could evolve into cancer. In this paper, we propose an algorithm that automatically identifies the abnormal cervical regions from colposcopy image sequence. Firstly, based on the segmentation of three different image regions, a set of low-level features is extracted to model the temporal changes in the cervix before and after applying acetic acid. Second, a support vector machine (SVM) classifier is trained and used to make predictions on new input feature vectors. As the low-level features are very insensitive to accurate image registration, only a rough normalization step is needed to sample image patches. Our preliminary results show that our algorithm is accurate and effective. Furthermore, our algorithm only needs to sample patches from six image frames within a five-minute time period. Hence, the proposed algorithm also could be applied to improve the accuracy of the mobile telemedicine for cervical cancer screening in low-resource settings.