Cervical Cancer Classification from Pap Smear Images Using Modified Fuzzy C Means, PCA, and KNN

Abstract

Cervical cancer is having the second-highest mortality rate next to breast cancer among women in developing countries. Early detection of the abnormality is the only way to prevent morbidity. As the decision about the abnormality of the cell is made manually by the traditional Pap smear test – the clinical test conducted for the detection of cervical cancer is more prone to false-negative and false-positive cases. This paper presents a novel approach for the automatic detection of cervical cancer using modified fuzzy C-means, extracting the geometrical and texture features, Principal Component Analysis (PCA), and classification. Modified fuzzy C-means show promising results in segmenting the input image into meaningful regions even when there is uncertainty. PCA is being performed to reduce the dimensionality of the data set by maintaining only the uncorrelated features thereby reducing the processing time of the algorithm. The classification of the pap smear images into normal and abnormal cells is being done by K Nearest Neighbour (KNN) classification with k-fold cross-validation and the result obtained in the proposed method is being compared with Fine Gaussian SVM, Ensemble Bagged trees, and Linear Discriminant. The efficiency of the proposed method is measured by calculating minimum accuracy, maximum accuracy, average accuracy, sensitivity, specificity, F1-score, and precision. The experimental results of the proposed method show impressive results with minimum accuracy 94.15%, maximum accuracy 96.28%, average accuracy 94.86%, sensitivity 97.96%, specificity 83.65%, F1-score 96.87%, and precision 96.31% for threefold cross-validation.