CerConvNet: Cervical Cancer Cells Prediction Using Convolutional Neural Networks

Abstract

Cervix cancer is a distinct form of cancer occurring in women, originating in the cells of the cervix, which is the region of the uterus connecting to the vagina. About 90% of cases of cervix cancer are related to human papillomavirus (HPV) infection. The mortality rate in developed nations has decreased because of routine HPV testing for women. The absence of reasonably priced healthcare facilities, however, continues to make it difficult for developing countries to offer inexpensive remedies. Therefore, developing an accurate algorithm for cervical cancer prediction is necessary to identify women who are at risk of developing this condition. Architectures of Deep Learning have been employed in recent years to construct accurate models for the prediction of cervical cancer. This study offers a unique, straightforward transfer learning framework: ResNet50, DenseNet201, EfficientNetb1 and InceptionResNetV2, to classify cervical images using the SIPaKMeD dataset and different performance measures are gathered and examined. Still, the recommended Densenet201outperformed the most advanced methods. We obtained an average accuracy of 98.78% with CNN models which is the highest compared over the existing models. Resnet50 achieved even better results after augmentation with an accuracy of 99.51% and Precision, Recall, F1-score of 0.99. As a result, the findings support our approach to providing low-cost first-level screening.