A Practical Segmentation Method for Automated Screening of Cervical Cytology

• Published in: 2011 International Conference on Intelligent Computation and Bio-Medical Instrumentation pp. 140 - 143
• Main Authors: Ling Zhang, Siping Chen, Tianfu Wang, Yan Chen, Shaoxiong Liu, Minghua Li
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2011
• Subjects: Algorithm design and analysis; Cancer; cervical cytology; Clustering algorithms; Color; Image color analysis; Image segmentation; nuclei segmentation; Shape; V channel
• ISBN: 9781457711527; 1457711524
• DOI: 10.1109/ICBMI.2011.4

In a full automatic cervical cytology screening process, one of the essential steps is the segmentation of cervical nuclei. Despite some progress, there is a need to improve sensitivity, speed, level of automation, and to reduce non-cellular artifacts. This paper presents a practical nuclei segmentation algorithm for solving these problems. The proposed approach first preprocess the V channel image from the HSV color space thus allowing accentuating the contrast between nuclei/leukocyte and cytoplast. In order to overcome the non-uniform illumination, the adaptive thresholding algorithm is utilized. Two characteristics named shape factor and roundness are introduced to validate if a segmented region is overlapped nuclei. Further, by exploring a concave-point based segmentation algorithm, overlapped even multi-overlapped nucleus can be separated. Experiment results carried out on 200 images (100 malignant and 100 normal) show that comparing with the past work [7], our approach can detect more malignant nuclei, less under-segmented normal nuclei, less debris/inflammatory cells and binarization error. Currently, our implementation on 1.9GHz dual-core computer takes 0.56s/image, on average. The proposed segmentation algorithm has potential application in full automated screening of cervical cytology. Furthermore, our algorithm shows promising performance when comparing with [14] on histopathological images.