Comparison of Popular CNN Architectures for Cervical Cancer Early Diagnosis

Cervical cancer has been considered by WHO as one of the deadliest diseases for women because of its high mortality and morbidity rates. It was reported that cervical cancer accounts for around 12% of all types of cancer in women worldwide. In fact, despite having very high morbidity and mortality rates, cervical cancer takes a relatively long time to mutate from normal cells to precancerous conditions. Therefore, it can be concluded that cervical cancer can be completely cured if it can be detected early. Therefore, research related to early detection of cervical precancerous cells is a must to prevent patient mortality and morbidity. This research is expected to be a method that can be used in diagnosing cervical cancer early. To achieve this goal, this study applies several architectures of the CNN algorithm such as MobileNetV2, RestNet50, InceptionV3, EfficientNetB0, Xception, and VGGNet16. As for the dataset for pattern recognition of cervical cancer, this study uses ThinPrep images which are considered as one of the best types of cervical cancer images. Based on the results obtained, of the 5 architectures used, this study succeeded in obtaining an accuracy value above 95%. Thus, it can be concluded that these 5 architectures are very feasible if applied to help early diagnosis of cervical cancer.