Deep CNN Models for Detecting Cervical Cancer in Pap Smear Images

Classifying the cervical cancer image dataset, which forms the basis of this research, is a crucial step in the initial screening for the presence of cancer cells in women. Classifying cervical cells in Pap smear images presents a significant challenge due to the inherent limitations of the images and the complex morphological variations in the structural characteristics of the cells. The detection of cervical cancer usually depends on conventional classification approaches that need feature extraction and cell segmentation procedures. However, in order to address problems with overfitting and poor performance with new data, Convolutional Neural Network (CNN) models necessitate extensive datasets. This project aims to develop a deep learning model for the automatic detection of cervical cancer, which operates without the need for segmentation techniques or specific features. The task involves classifying four classes using transfer learning techniques, which leverage knowledge and skills gained from previous tasks for new ones. This research assessed and compared 15 CNN models using the publicly available RepomedUNM dataset to classify cervical cancer cells. From the evaluation results, two models with the best performance (above 90% accuracy) were found, namely Xception and InceptionResNetV2.