Cervical Cancer Detection Using Deep Transfer Learning Models: A Comparative Study

Cervical cancer is an insidious disease that emerges in the cervix, the narrow passage between the uterus and vagina, and is primarily triggered by the human papillomavirus (HPV), which is a widespread sexually transmitted infection. Cervical cancer risk factors include age and hormonal contraceptive use, among others. Timely detection of cervical cancer is crucial in improving survival rates and minimizing mortality rates. In recent times, deep learning algorithms have become valuable tools for analyzing and diagnosing medical images. New and advanced research has clearly shown that using advanced deep learning algorithms to automatically assess high-resolution images of the cervix is a highly effective way to improve cervical cancer screening with visual inspection using acetic acid (VIA). This approach can make cervical cancer screening more accurate and reliable. A dataset consisting of 25,000 images distributed into three classes (training, testing, and validation) was utilized in our study. We did our research carefully, using six different deep learning models: VGG-16, VGG-19, MobileNet, EfficientNet, InceptionV3, and ResNet50. 3 of our experiment using VGG-16, VGG-19, and MobileNet achieved an accuracy of 99.00%. Upon analyzing our results in comparison to prior research, we discovered that our transfer learning models exhibited superior efficacy in detecting cervical cancer based on specific evaluation metrics.