Design And Implementation Of Medical Image Retrieval System

With the rapid development of medical imaging technology and the popularity of each big hospital information network, available for clinical and research use of medical image is rapidly expanding. Nowadays, medical imaging equipment such as CT,MRI,PET each day will produce about 10 g to 20 g medical image data.In the face of vast amounts of medical image collections, how to quickly and accurately find the image we need from the huge image repository, how to extract valuable information to assist doctors diagnosis and easy to scientific research and teaching has already become a hot spot in the field of medical image on the computer over the years. Therefor, the image retrieval technology is applied to medical field has great significance.In recent years, researchers at home and abroad to carry out a large number of studies of medical image retrieval technology, made a certain progress, in some foreign mature medical image retrieval system has been successfully applied to clinical diagnosis and treatment of scientific research and teaching. At home, text based medical image retrieval has been widely used, but there is still no very mature content-based medical image retrieval system practical application in clinical and put the content-based medical image retrieval technology into the PACS system also cannot be achieved in a short time.This paper studied the medical image retrieval technology, made detail comparative analysis about the feature extraction according to the features of medical images, design a based on the combination of text and content of medical image retrieval system. Main research work are as follows.(1) Study the DICOM medical image automatic text information extraction technology. Implement the patient ID, Image study date, image modality in DICOM file as text features for text information retrieval.(2) Study the multiple feature fusion based on content retrieval technology, select the texture feature and shape feature for image to used to retrieve the underlying features. Based on the algorithm’s advantage complementary, to make similar features for normalization, then compare the search results of the texture and shape feature weighting fusion.(3) Study model of the multistage retrieval. According to the different needs of users, the system designed by this paper can detect two levels retrieval. Primary retrieval based on text retrieval, use DICOM text information matching the images in the library for preliminary screening. Secondary retrieval based on content features fusion retrieval, retrieval similar images through the underlying features.At last, this paper uses the CT images from different parts of different patients, draw the following conclusion, more features fusion technology can effectively use different advantages, it can more fully express the visual content of the image, the retrieval effect is better than a single feature of retrieval results. Weight distribution of features also has a certain effect on search results, the higher weights of texture feature, the retrieval results will be better. The combination of text and content retrieval methods can inherit the advantages of both, improve the retrieval accuracy with feasibility and practicability.