Comparative Study Between Sparse Representation Classification and Classical Classifiers on Cervical Cancer Cell Images

Abstract: The pap-smear classification is still a challenging task for machine–aided cervix cancer diagnosis because it is tedious even for a trained cytologist to analyse and diagnose each slide obtained from different patients. Various methods have been proposed for medical image classification. In this paper, a multi-feature set sparse representation classification (mfSRC) is proposed in order to classify the pap smear cell images. Here, this representation goes through a training stage employing Genetic Algorithm guided by multi feature set dictionary learning approach. The data consists of 917 images of Pap-smear cells, classified carefully by cyto-technicians and doctors. Each cell is described by 20 numerical features, and the cells fall into 7 classes. In order to understand the relevance of sparse representation in the classification of pap smear images, the performance of other classical classifiers were also evaluated. Results show that classification accuracy of sparse representation generally outperforms other classical classifiers.

Keywords: Cervical cancer, Sparse Representation, Classical classifiers, Genetic Algorithm, Pap smear