A Bi-Stage Technique for Segmenting Cervical Smear Images Using Possibilistic Fuzzy C-Means and Mathematical Morphology

Accurate detection of cervical cells in microscopic smear images is an integral part of image-based efforts for cervical cancer diagnosis. The study presents a bi-stage technique to segment smeared cervical (CS) images. In the first stage, the Possibilistic Fuzzy C-Means (PFCM) algorithm, an appendage to the standard FCM algorithm that was developed mainly to address some weaknesses of the standard Fuzzy C-Means (FCM) algorithm in terms of its poor sensitivity to noise, is employed in order to segment the cervical cell into clusters. Following this, in the second stage, the cervical cell components, i.e., nucleus and cytoplasm, are detected and delineated using the mathematical morphological operations. Using a dataset of cervical smear images, the proposed technique achieves average values of 0.98, 0.98, 0.97, and 0.93 for sensitivity, specificity, accuracy, and Zijdenbos Similarity Index (ZSI) respectively, for the segmented nucleus. Similarly, for the segmented cytoplasm values of 0.89, 0.96, 0.91, and 0.95, respectively are obtained for the same parameters. These experimental results suggest that the PFCM algorithm obtains higher average ZSI values than the standard FCM algorithm for both the segmented nucleus and cytoplasm indicates the potential application of the proposed study in cervical cancer diagnosis.