Cervical Cell Classification using Fourier Transform

Abstract

A method of classifying the precancerous cells from Papanicolaou smear images is proposed in this paper. The proposed method utilizes a set of simple features extracted from the two-dimensional Fourier transform of the cell images in order to avoid the problem of cell and nucleus segmentation. The features used to discriminate between the normal and the abnormal cells are calculated based on the mean, variance, and entropy obtained from the frequency components along the circle of radius r centered at the center of the spectrum and the frequency components along the radial line having an angle θ. The classification results achieved by five classifiers are compared in order to evaluate the utilization of the selected features in normal and abnormal cell classification using fourfold cross validation. The classifiers used in this research include Bayesian classifier, linear discriminant analysis (LDA), K-nearest neighbor (KNN) algorithms, artificial neural network (ANN), and support vector machine (SVM). The classification rates obtained from these classifiers show promising performances. The result from the support vector machine provides the best accuracy and the lowest false rate. It achieves more than 92% correct classification rate on a set of 276 cervical single-cell images containing 138 normal cells and 138 abnormal cells.