Performance Comparison of Convolutional Neural Network Models on Cervical Cell Classification
Syed Mohd Zahid Syed Zainal Ariffin, Hazwani Hasnun Ali Musa, Nursuriati Jamil · 2021
Cervical cancer is regarded as one of the most common type of cancer suffered by women in the world. However, it can be treated if being detected early. The main method of diagnosing this type of cancer is through Papanicolaou test, in which cervical cells are collected and examined under a microscope. The process of examining the cervical cells is often done manually and is labourious. Several methods have been proposed to automate this process using machine learning. This study explores three Convolutional Neural Network models (i.e., AlexNet, Inception-V3 and ResNet50-V2) in classifying cervical cells. Transfer learning was implemented with different sets of varying number of images. Herlev dataset was used in this study. In the Herlev dataset, images were classified into seven classes (i.e. four abnormal and three normal). The classification done was based on seven-class classification. Performance of each model was measured based on accuracy, precision and recall. Based on the result obtained, ResNet50-V2 performed the best with the accuracy, precision and recall compared to the other two models.