Research On Content-based Medical Image Retrieval

Along with the rapid development of medical imaging technology such as X radial,CT,MRI and so on, medical images become more digital and more various. The application of PACS system results in lots of digital medical images stored in the computer. These images enrich researcher's research and teaching, but it's also a difficult problem for the researchers to manage the great number of images. Traditional text-based medical image retrieval has a lot of disadvantages, which depends on personal idea too much and can't reflect the primitive characteristic of the image, so this method is more and more out of date. In this situation, content-based medical image retrieval develops quickly.Extracting the characteristic of image is one of the key technology to content-based medical image retrieval. In fact, medical images have some particular characteristic. We review adequately current methods of extracting the features of images, and then combine the imaging mechanism and special characters of medical images to research the suitable algorithms. Algorithms discussed mainly as follows: gray character, including gray histogram and Cumulative Distribution Function; texture character, including gray co-occurrence matrix and gray level-gradient co-occurrence matrix; shape character, including Hu moment and Zernike moment. All of these, we give the experimental results, and discuss the effectivity of the results by using the precision percentage and recall percentage. Then we combine the cumulative distribution function, CDF and gray level-gradient co-occurrence matrix to retrieve the medical images, which has been proved a good performance.Generally speaking, single character can not describe the image totally. Gray character is good for the distribution of gray of image, but it can't reflect the spatial information of image. Similar distribution of gray may be a very different shape character. To solve this problem, we present a matrix of regional location to retrieve the medical images. This method considers the relationship of both gray and space. The result shows that this method has good performance.