Segmentation of Cervical Cell Nucleus using Intersecting Cortical Model Optimized by Particle Swarm Optimization

Changes in the morphology of cervical cell nucleus are one of the most important features to be observed during Pap-smear screening. In this study, Intersecting Cortical Model (ICM) was employed to segment the nucleus from cervical cell images. The four unknown parameters in ICM were optimized by Particle Swarm Optimization (PSO). Two hundred and fifty test images were randomly selected from Herlev dataset. The segmented results were compared with Otsu thresholding, Expectation Maximization technique, region growing and Fuzzy C-Means clustering technique. Analyses revealed that ICM produced the best segmentation result, with Zijdenbos Similarity Index (ZSI) of 0.914, Peak Signal to Noise Ratio (PSNR) of 62.946 dB, Misclassification Error (ME) of 0.056 and Relative Foreground Area Error (RAE) of 0.132. Wilcoxon Signed-rank Test reported ICM significantly outperformed the four comparison techniques, with p-values less than 0.05 for all the performance metrics.