Domain-based analysis of colon polyp in CT Colonography using image-processing techniques

Document Type

Article

Publication Title

Asian Pacific Journal of Cancer Prevention

Abstract

Background: The purpose of the research was to improve the polyp detection accuracy in CT Colonography (CTC) through effective colon segmentation, removal of tagged fecal matter through Electronic Cleansing (EC), and measuring the smaller polyps. Methods: An improved method of boundary-based semi-automatic colon segmentation with the knowledge of colon distension, an adaptive multistep method for the virtual cleansing of segmented colon based on the knowledge of Hounsfield Units, and an automated method of smaller polyp measurement using skeletonization technique have been implemented. Results: The techniques were evaluated on 40 CTC dataset. The segmentation method was able to delineate the colon wall accurately. The submerged colonic structures were preserved without soft tissue erosion, pseudo enhanced voxels were corrected, and the air-contrast layer was removed without losing the adjacent tissues. The smaller polyp of size less than < 10mm was detected correctly. The results were statistically validated qualitatively and quantitatively. Segmented colons were validated through volumetric overlap computation, and accuracy of 95.826±0.6854% was achieved. In polyp measurement, the paired t-test method was applied to compare the difference with ground truth and at α=5%, t=0.9937 and p=0.098 was achieved. The statistical values of TPR=90%, TNR=82.3% and accuracy=88.31% were achieved. Conclusion: An automated system of polyp measurement has been developed starting from colon segmentation to improve the existing CTC solutions. The analysis of domain-based approach of polyp has given good results. A prototype software, which can be used as a low-cost polyp diagnosis tool, has been developed.

First Page

629

Last Page

637

DOI

10.31557/APJCP.2019.20.2.629

Publication Date

2-1-2019

This document is currently not available here.

Share

COinS