A Review of Image-Based Deep Learning Algorithms for Cervical Cancer Screening

Abstract

The significant advance in artificial intelligence has posed many challenges, with disease detection being one of the most important. Early detection can be very important in preventing progressive disease progression and can help provide accurate treatment options. Cervical cancer is the fourth type of cancer most common in women. In 2018, 570 000 cases were estimated in women around the world. This article aims to present a review of different image-based algorithms for cervical cancer screening. For the research process, three important sources of information were considered: Scopus, Web of Science, and PubMed, considering a total of 12 articles taking into account the last five years. The articles were analyzed considering the databases used, the preprocessing of the images, the segmentation of the images, the classification of images, and the proposals' results. The results show great advances in the techniques used for cervical cancer screening, with convolutional neural networks being the most widely used technique. In addition, including the segmentation stage in the construction of the models can significantly increase precision. Finally, it is shown that the k-fold cross validation technique is one of the most used and efficient techniques to validate the models.