Clustering Techniques on Pap-smear Images for the Detection of Cervical Cancer

Abstract

A Pap smear test is the most efficient and prominent method for the detection of dysplasia in cervical cells. Pap smear is time-consuming and sometimes it is an erroneous method. Computer-assisted screening can be widely used for cervical cancer diagnosis and treatment. Most of the existing approaches do not give good performance on real images due to poor staining, dye, blood and inflammatory cells. In our proposed approach, we are extracting nucleus only from the Pap smear images. For segmentation Laplacian of Gaussian (LOG) filter and morphological operations has used for edge detection. In the classification phase, two clustering techniques K-means and Fuzzy c-means (FCM) has been used using Principle Component Analysis (PCA). The classification of Pap smear images is based on the Bethesda System. The approach has performed on a dataset obtained from pathologic lab containing 40 Pap smear images with 500 cells. Performance evaluation has done using Purity and Jaccard Index (JI). The purity of K-means is 0.815 and for FCM it's 0.875.