Multilayer Semantic Retrieval Facing To Charaterization Of Lung CT Image

With the development of computer and medical image technology, medical image hasbecome an important basis to diagnose and analysis diseases for the doctor. Modern hospitalproduces large amounts of medical image data every day, one of the problem to be solvedwhich doctors facing is how to find specific images in the vast amounts of data quickly andaccurately. Combine the semantic-based image retrieval technology and specialized fieldknowledge of medical images characterization on the basis of analyzing a variety of imageretrieval algorithm. A semantic-based medical image retrieval system has been designed.The brief review of the history of the development of retrieval technology is given in thebeginning of the article. Research status in semantic medical image retrieval based researchboth at home and abroad is analyzed. The key technology of semantic retrieval is introducted.The through research in lung CT image is conducted. Thus, a lung parenchyma segmentationalgorithm is proposed for lung CT image. The algorithm is easy to implement and has a betterexperimental results. Use conventional method to extract the rough contour of pulmonaryparenchyma. In connection with the absence of lung parenchymal lesions in the previous step,an improved two-dimensional convex hull algorithm is proposed to repair the pulmonaryparenchyma contour. Acquire the pulmonary parenchyma internal contour by using regionalgrowth and morphology comprehensively. Firstly, an improved two-dimensional convex hullalgorithm is proposed for extracting the meaningful regions and the lung parenchyma isobtained. Secondly, gray and texture characteristics are extracted both in the original CTimage and the lung parenchyma image. The features are gray histogram statisticalcharacteristics, gray level co-occurrence matrix characteristics, Tamura texture feature andGabor wavelet texture feature. These characteristics of the original image as the globalfeatures of the image, while the divided lung parenchyma of these characteristics of the imageas a local feature. Thirdly, the high-level semantic features of the image are analyzed.Semantic keywords are proposed by statistics the high frequency vocabulary which candescribe the image characterization given by doctors. Return a total of six key words(cavityand hollow, nodular and lumps, water arc low density, blotchy high density film, patchy highdensity film patchy and plaque high density film). A semantic level model for medical imageis proposed. Then, semantic mapping from low-level features to high-level semantic featuresis realized by using k-nearest neighbor algorithm. Experimental results show that thealgorithm has a good mapping result, and it is a simple and effective method of semanticmapping. Lastly, by using manifold learning theory, a feedback method based on image manifold is used in order to meet the user’s retrieval intention.According to the semantic level model for medical image and associated algorithmproposed in this paper, by using Visual C++platform and My SQL database a medical imageretrieval system based on semantic is designed and implemented. Experiment results showthat the system can achieve the semantic keywords retrieval for lung CT imagecharacterization.