Automated cervical cancer detection using pap smear images

Abstract

Cervical cancer is the most common cancer among the women. Pap smear screening is the most effective test for detecting the cervical precancerous. But this process requires a long time to complete and also may be an erroneous procedure. In this paper, an automated cervical cancer detection method is presented. This method introduces adaptive median filter to remove impulse noises from the Pap smear images and then uses bi-group enhancer to discriminate the nuclei pixels from other object pixels. Then, segmentation methodology is presented to separate the nucleus regions from the cervical smear images. Two clustering-based classifiers, minimum distance and K-nearest neighbor classifiers, have been used in the classification phase for verifying the performance. The technique was evaluated using 158 Pap smear images from DTU/HERLEV Pap smear benchmark database. The accuracy of the detection method is 92.37 and 98.31 % for minimum distance and K-nearest neighbor classifiers, respectively.