期刊名称：International Journal of Mechatronics, Electrical and Computer Technology
印刷版ISSN：2305-0543
出版年度：2017
卷号：7
期号：26
页码：3644-3652
出版社：Austrian E-Journals of Universal Scientific Organization
摘要：With the pervasive availability and application of medical imaging equipment, every day, thousands of digital medical images must be stored in databases of hospitals and other medical facilities. Considering the sheer number of medical images taken every day, organization, classification, and retrieval of these images has become very challenging. One of the typical methods of image retrieval is the use of bag of visual words. In this approach, one of the most common techniques of constructing visual words is the use of K-means clustering. However, performance of K-means clustering largely depends on its initial cluster centers, and given the randomness of cluster center initialization in K-means, those BoW- based methods that utilize basic K-means often show a substandard performance. This paper presents a method to improve the accuracy of content-based medical image retrieval by optimizing the initial cluster centers of K-means by an ensemble clustering approach.In this method, after extracting the local features by SIFT descriptor, K-means clustering with random initialization will be run several times to yield several base clustering, and then a consensus algorithm selects, from among these initial results, the best clustering and more specifically the best visual words in terms of clustering criteria. This technique was tested on a group of standard x-ray images and the results showed that the use of ensemble clustering with MCLA serving as consensus algorithm leads to visual words that are more efficient and accurate than those created with randomly initialized centers.
关键词：content-based medical image retrieval;bag of words; SIFT descriptor;ensemble clustering.
