Cervical Cell Classification using Learning Vector Quantization (LVQ) Based on Shape and Statistical Features

Abstract

This research was conducted to design a system that is able to classify cervical cells into two classes, namely normal cells or abnormal cells. We use digital images of single cervical as research materials and Learning Vector Quantization (LVQ) as classification method.  Prior to classification, the nucleus areas of single cervical cell images were segmented and features were extracted. The features used in this study are 7 kinds of which consist of 2 types of feature, namely shape features and statistical features. The shape features used are area, perimeter, shape factor, and roundness of the nucleus, while the statistical features of the grayscale image histogram used are mean, standard deviation, and entropy. LVQ optimal parameter values based on the highest accuracy of training data, are learning rate 0.1 and learning rate reduction 0.5. The highest accuracy of system obtained from 45 testing data is 93.33%.