Automatic Cervical Cell Classification Using Patch-Based Fuzzy Clustering and Minimum Average Correlation Energy Filter

Abstract

Cell segmentation and cell classification are important steps in an automatic cervical cell detection system. In this paper, we propose a method to automatically segment and classify cervical cells. A nucleus is segmented using the patchbased fuzzy c-means (FCM) algorithm. Each segmented nucleus is classified as normal or abnormal using the minimum average correlation energy (MACE) filter. Low-frequency components of the MACE filter are discarded in order to reduce the complexity of computation. The accuracy of segmentation by the patch-based FCM is compared to the segmentation by the active contour model (ACM). The segmentation performance is evaluated by the probability of error and the Hausdorff distance comparing to the manual segmentation by an expert. The nucleus segmentation by using the patchbased FCM clustering method corresponds well with the expert’s opinion and yields better results than the ACM. The overall segmentation error by the patch-based FCM is approximately 6%. Four-fold cross validation is performed to test the efficiency of the proposed MACE filter on both ACMand FCM-based segmented images. Using the patch-based FCM segmentation together with the compact MACE filter yields a good classification rate of about 85% which is higher than that using ACM-based segmentation. The proposed segmentation and classification methods yield promising results for further development in an automatic cervical cell detection system.