Diagnosis of cervical cancer using texture and morphological features in pap smear images

Abstract

Background: Cervical cancer is one of the most common cancers among women worldwide, which can be diagnosed more quickly via using digital systems. The purpose of this study was to classify the cells in Pap smear test images into two types of normal and abnormal by using image processing to diagnose cervical cancers. Methods: We used Herlev public database, which contained 917 cells. 35 geometric and 263 histologic features such as Gray Level Co-Occurrence Matrix (GLCM), Local Binary Pattern (LBP), and rotational gradient histogram were extracted from cell images. T test filter method was applied on the data set after extraction of geometrical and textural features. We used different classification methods such as support vector machine (SVM), decision tree (DT), k nearest neighbor (KNN) and ensemble classifiers. Findings: The best results were for SVM classifier as 97.5 accuracy in two-class classification with 20 features. Conclusion: Feature selection and feature extraction methods are very important for classify normal and abnormal cervical cell images. By optimizing and choosing the right methods, we can optimizing accuracy, and speed and error (2-3 percent). © 2020 Isfahan University of Medical Sciences(IUMS). All rights reserved.