Abstract—Early detection of cervical cancer involves visual screening for changes in cellular morphology through microscopic analysis of Pap smears. Cytological interpretation by conventional microscopy of abnormal Pap smears performed manually is time-consuming, observer dependent and error prone. The aim of this study is to discriminate abnormal squamous cells from normal ones by quantitative image analysis of cervico-vaginal single cells with specific focus on the structure of the nuclei. In this study: 1) Six discriminative features such as nuclear area, nuclear perimeter, equivalent diameter, major axis length, minor axis length and convex area were selected and statistically justified, 2) A new dataset of 100 Pap smear cell images were collected from North-East Indian Regional population for the experimentation, and 3) Ground truth images of Pap smear cell dataset created by medical experts were compared with the automatically-segmented images with respect to the selected shape features. The cell boundary was segmented using greedy active contour model. Based on these six discriminating features, relevant cell images were classified as normal and abnormal using Support Vector Machine. Our method reports accuracy of 97.33%. Additionally, the proposed framework was applied to a known Pap smear benchmark dataset, to which we report an accuracy of 90.21%.

Index Terms—cervical cancer, cell imaging, pap smear test, nucleus, segmentation, greedy active contour model, feature extraction, support vector machine

Cite: Mrinal Kanti Bhowmik, Niharika Nath, Abhijit Datta, and Anjan Kumar Ghosh, "Shape Feature Based Automatic Abnormality," Journal of Image and Graphics, Vol. 5, No. 2, pp. 52-58, December 2017. doi: 10.18178/joig.5.2.52-58