Automated System For Efficient Cervical Cancer Screening Using Cervicogram Images

Cervicography, a diagnostic technique that checks and identifies problems in the cervix area with its major aim being cervical cancer, is used. This involves application of 5% acetic acid to the cervix followed by capturing detailed images of it through special imaging technology. The proposed method offers an alternative to conventional ones such as Pap smears that are both time-consuming and call for well trained staff. Cervicography is widely used in community screening programmes due to its ability to be completed quickly and efficiently on a larger number of women. Furthermore, it has improved sensitivity for detecting abnormalities in the cervical canal thus providing clearer pictures of the cervix while also increasing possibilities for finding premalignant or malignant growths. The purpose of this study is to use cervicogram images for improvement of the cervical cancer screening techniques. This study proposes a Dirichlet Ensemble Model with RoI extraction of cervix using You Only Look Once (YOLO) framework. The classification model uses Kaggle Intel and MobileODT dataset for training, validation and testing. The testing accuracy obtained as 80%, F1 score of 0.79 and ROC AUC score as 0.8327.