Color canals modification with canny edge detection and morphological reconstruction for cell nucleus segmentation and area measurement in normal Pap smear images

Abstract

This paper presents a cell nucleus segmentation and area measurement of Pap smear images by means of modification of color canals with Canny edge detection and morphological reconstruction methods. Cell nucleus characterization plays an important role for classifying the degree of abnormality in cervical cancer. The aim of this work is to find the matched measurement method with the manual nucleus area measurement. In this work, we utilized pap smear single cell images from Herlev data bank in RGB mode. The cell images were selected from 90 normal class subjects that include: Normal Superficial, Normal Intermediate, and Normal Columnar classes. The nucleus of each cell image was cropped manually to localize from the cytoplasm. The color canals modification was performed on each cropped nucleus image by, first, separating each R, G, B, and grayscale canals, then implementing addition operation based on color canals (R+G+B, R+G, R+B, G+B, and grayscale). The Canny edge detection was applied on those modifications resulting in binary edge images. The nucleus segmentation was implemented on the edge images by performing region filling based on morphological reconstruction. The area property was calculated based on the segmented nucleus area. The nucleus area from the proposed method was verified to the existing manual measurement (ground truth) of the Herlev data bank. Based on thorough observation upon the selected color canals and Canny edge detection. It can be concluded that Canny edge detection with R+G+B canal is the most significant for all Normal classes (r 0,305, p-value 0.05). While for Normal Superficial and Normal Intermediate, Canny edge detection is significant for all RGB modifications with (r 0.414 - 0.817 range, , p-value 0.05), and for Normal Columnar, Canny edge detection is significant for R+B canal (r 0.505, p-value 0.05).