Data Augmentation Techniques to Detect Cervical Cancer Using Deep Learning: A Systematic Review

Computer-assisted systems have been widely used as tools to support medical experts in various fields, including the analysis of cervical cytology. However, due to patient privacy and ethical considerations, processing the model is challenging due to insufficient data in medical imaging. Data augmentation has gained popularity as a solution to this problem, especially in sectors where large datasets are unavailable, thereby increasing the size of a training dataset. This study aimed to identify a data augmentation technique to detect cervical cancer. To conduct this analysis, we systematically reviewed secondary studies published between 2017 and 2023 using the search term 'data augmentation', 'cervical cancer' and 'deep learning' from databases including Scopus, Web of Science, PubMed, IEEEXplore, Science Direct, and conducted a manual search on Google Scholar. The results showed that data augmentation techniques are categorized as basic methods and artificial image generation. Among basic data augmentation techniques, rotation and flipping are the most widely used. In the generation of artificial images, DCGAN is used to create high-quality synthetic images. Basic augmentation is used for both segmentation and classification tasks, while artificially generated techniques are used exclusively for classification tasks. Consequently, all these techniques enhance the performance and generalizability of deep learning models by increasing the size of the dataset.