Title: Cooperative pixel clustering for accurate automatic inflamed appendix extraction from ultrasound images
Authors: Kwang Baek Kim; Doo Heon Song; Hyun Jun Park
Addresses: Department of Artificial Intelligence, Silla University, Busan 46958, Korea ' Department of Computer Games, Yongin Songdam College, Yongin 17145, Korea ' Division of Software Convergence, Cheongju University, Cheongju 28503, Korea
Abstract: Reliable diagnosis and management of acute appendicitis is a difficult problem. Automatic extraction of inflamed appendix from ultrasonography is desirable to minimise the operator subjectivity of the ultrasound image analysis. In this paper, we propose a cooperative unsupervised machine learning approach to this automatic segmentation problem. The quantisation process is done by fuzzy ART with dynamic controlled vigilance parameter and fuzzy C-means pixel clustering with good parameter initialisation related with fuzzy ART. Two results are combined to produce a conservative but reliable inflamed appendix object formation. In the experiment using 80 DICOM format ultrasonographic images with inflamed appendix, the proposed method was successful in 77 cases or 96.25% correct by pathologists' evaluation which is much better performance than previous edge detection-based approach whose performance was less than 83%. This new approach is also relatively immune to the appendix shape which was a weak point of previous pixel clustering approaches.
Keywords: appendicitis; ultrasound; fuzzy ART; fuzzy C-means; FCM; image quantisation.
DOI: 10.1504/IJCVR.2021.118536
International Journal of Computational Vision and Robotics, 2021 Vol.11 No.6, pp.640 - 652
Received: 29 Oct 2019
Accepted: 14 Aug 2020
Published online: 28 Oct 2021 *