Automated Cervical Cancer Detection from Digital Cervical Cytology Images Using Deep Learning

Abstract

Cervical cancer is a major health problem in the world mainly due to limited accuracy in manual screening using manual cytology test and late diagnosis. An automatic diagnosis system of cervical cytology image could help overcome these issues. A novel automatic system of cervical cancer detection using deep learning and digital cervical cytology image is proposed in this study. A convolutional neural network (CNN) architecture, combined with conventional image preprocessing and data augmentation methods, is used to learn features from the images. It is then trained on a publicly available dataset which is the Herlev Pap smear database. In the database, images are divided into normal and abnormal classes. It achieves the classification accuracy of 96.1%, the F1-score of 96.1%, and the ROC-AUC of 0.97. The experimental results, compared to recent state-of-the-art methods, demonstrate that the proposed CNN method significantly outperform current deep learning approaches. The proposed system can potentially act as a helpful computer-aided diagnosis tool for cervical cancer detection.