Cervical Cancer Diagnosis Using Time-Lapsed Colposcopic Images

Abstract: Cervical cancer is the fourth top cancer-related deaths of women worldwide. Discovery of cervical intraepithelial neoplasia (CIN) in the initial stage can rise the existence rate of the patients. The structures of the unique (pre-acetic-acid) image and the colposcopic images captured at around 60s, 90s, 120s and 150s during the acetic acid test are fixed by the feature instruction networks. we recommend a deep learning framework for the exact identification of LSIL+ (including CIN and cervical cancer) using time-lapsed colposcopic images. The projected framework includes two main mechanisms, i.e., key-frame feature encoding networks and feature fusion network. Some fusion approaches are associated, all of which outstrip the remaining automated cervical cancer diagnosis systems using a particular time slot. A graph convolutional network with superiority features (E-GCN) is initiate to be the greatest appropriate fusion approach in our study, due to its outstanding explain ability consistent with the clinical preparation. A large-scale dataset, covering time-lapsed colposcopic images from 7,668 patients, is collected from the collective hospital to train and confirm the deep learning framework. Colposcopists are enquired to contend with our computer-aided diagnosis system. The proposed deep learning framework understands a classification precision of 78.33%—similar to that of an in-service colposcopist—which confirms its possible to transport assistance in the realistic clinical situation.

Keywords: Cervical cancer, acetic acid test, graph convolutional network, feature fusion.