Research And Application Of Content Based Medical Image Retrieval

The explosive growth of medical image data and the rapid development of data mining technology have put forward new requirements to the medical image retrieval technology. However, the performance of traditional image retrieval technology in medical image retrieval is not always satisfactory. In order to apply the technology of image retrieval to the field of Computer Aided Diagnosis(CAD), medical teaching practice and medical scientific research, this thesis has made some research on image retrieval technology and characteristics of medical images, and then proposed an improved scheme.Firstly, this thesis studied the key technologies of image retrieval and the characteristics of digital medical image. After analyzing the research status of image retrieval technology at home and abroad, this thesis focused on the content based image retrieval technology. Based on the research of image retrieval system in image feature extraction, feature matching, rele vance feedback and evaluation of retrieval performance, this thesis put forward the application of image retrieval technology in the specific field of medical image.By comparing the commonly used color features and texture features in medical image retrieval, the conclusion was drawn that the texture feature is more conducive to medical image retrieval. In combination with the limitations of medical image segmentation and the characteristics of medical image, the Gray Level Co-occurrence Matrix(GLCM) was selected as the global feature of medical image retrieval. And an improved scheme of GLCM, Weighted Sub-block Gray Level Co-occurrence Matrix(WS-GLCM) was proposed. According to the performance of the local invariant feature of SIFT in medical image retrieval, it was selected as the local feature of medical image retrieval. Based on the research of Bag Of Words(BOW) model, then WS-GLCM and SIFT was combined, two improved schemes were proposed, namely, the SIFT feature with global texture context and the SIFT feature with local texture context.In view of the limitations of Support Vector Machine(SVM) based relevance feedback technology in medical image retrieval. On the basis of the study of SVM classifier and the algorithm of Bootstrap aggregating(Bagging) classifier, a new improved scheme based on SVM for relevant feedback was proposed. The problems of insufficient training samples in the artificial feedback module, and the low accuracy of the SVM classifier for medical image classification were then effectively solved.Finally, the application of medical image retrieval technology in CAD, medical practice and medical research fields were briefly discussed. In view of the present situation of storing and using medical information data and medical image data, and the rapid development of the technology of machine learning and cloud computing, the main development directions of the future medical image retrieval technology were proposed. And a good prospect was made that digging huge medical value from the medical data and then promoting the improvement of medical level.