Classification of Cervical Cancer Detection using Machine Learning Algorithms

Abstract

Cervical cancer is one of the diseases which are most prevalent in females. This is seen when some changes occur in a woman's cervix. To further complicate the issue these cancer cells can spread to other organs such as the liver, bladder, rectum, and even lungs. Earlier detection, screening, and careful precautions have seen higher rates of recovery. Colposcopy and pap-test are two methods to obtain the cells and study the subject. The pap-smear test is more preferred over Colposcopy as it's pain-free, low-cost, and has a more distant range. A proper cell image acquisition and accurate segmentation are the key steps for the correct classification of cells. This article presents a Machine Learning (ML) classification method applied to the Herlev pap-smear image Dataset, using Support Vector Machines (SVMs). For the segmentation phase, active contour models that are driven using Gaussian Fitting Energy were utilized. These segmented images were compared with the manually annotated images that were provided by professional cytologists. The dice index was the parameter used from the comparison, which reported a match of 92%. With polynomial SVMs highest classification accuracy was 95%.