Automatic detection of cervical cells in Pap-smear images using polar transform and k-means segmentation

Abstract

We introduce a novel method of cell detection and segmentation based on a polar transformation. The method assumes that the seed point of each candidate is placed inside the nucleus. The polar representation, built around the seed, is segmented using k-means clustering into one candidate-nucleus cluster, one candidate-cytoplasm cluster and up to three miscellaneous clusters, representing background or surrounding objects that are not part of the candidate cell. For assessing the natural number of clusters, the silhouette method is used. In the segmented polar representation, a number of parameters can be conveniently observed and evaluated as fuzzy memberships to the non-cell class, out of which the final decision can be determined. We tested this method on the notoriously difficult Pap-smear images and report results for a database of approximately 20000 patches.