Deep Feature Extraction for Pap-Smear Image Classification

Abstract

Cervical cancer is one of the major public health problems in the world. Even if it is one of the most preventable cancers, the early screening with reliable Pap-smear test is the key to curing. Based on the morphology and the texture of the cervix cells, Cytopathologists rely on hand-crafted features to determine whether a cellule is normal or abnormal. In medical imaging, deep learning can generate features that are more sophisticated and difficult to elaborate in descriptive means. These relevant features allow improving the classification's accuracy. In this study, we use deep neural networks to extract features from Pap-smear images and provide these extracted features as inputs for optimized MLP classifier. We study the ability of the four different pre-trained models as feature extractors toward classifying Pap-smear images. Experiments performed on the DTU/HERLEV Database and comparison show that for the Pap-smear image classification, the ResNet50 exceeds the VGGs and the InceptionV3 by 15% of accuracy.