RepoMedUNM: A New Dataset for Feature Extraction and Training of Deep Learning Network for Classification of Pap Smear Images

Morphological changes in the cell structure in Pap Smear images are the basis for classification in pathology. Identification of this classification is a challenge because of the complexity of Pap Smear images caused by changes in cell morphology. This procedure is very important because it provides basic information for detecting cancerous or precancerous lesions. To help advance research in this area, we present the RepoMedUNM Pap smear image database consisting of non-ThinPrep (nTP) Pap test images and ThinPrep (TP) Pap test images. It is common for research groups to have their image datasets. This need is driven by the fact that established datasets are not publicly accessible. The purpose of this study is to present the RepoMedUNM dataset analysis performed for texture feature cells on new images consisting of four classes, normal, L-Sil, H-Sil, and Koilocyt with K-means segmentation. Evaluation of model classification using reuse pretrained network method. Convolutional Neural Network (CNN) implements the pre-trained CNN VGG16, VGG19, and ResNet50 models for the classification of three groups namely TP, nTP and all datasets. The results of feature cells and classification can be used as a reference for the evaluation of future classification techniques.