Retrieval Technology Research Of Calcification Lesions In Mammograms

Breast cancer is a common form of malignant tumors which threaten women health; there is an increasing trend in its incidence recent years. It shows clearly that early discovery; early diagnosis and early treatment of breast cancer can significantly increase the chance of survival and reduce the mortality of breast cancer for patients. Mammography has become the most common mean because of its convenience and efficiency in the current breast cancer screening. The micro-calcifications which are small and granular in mammograms are an important early sign for early breast cancer. One of the key technologies is to find micro-calcifications in mammograms in time and judge whether they will become malignant or not in early diagnosis. The system of content-based image retrieval (CBIR) not only eliminate the need for the burden for the design and training of classifier, but also make full use of the original data which the calcifications already has been diagnosed, thus it can help physicians to improve the diagnostic accuracy effectively.In this dissertation, on the basis of analyzing the key technologies and trends about the content-based image retrieval, we conduct a systematic, depth and comprehensive study on the key issues which the technology of content-based medical image retrieval faced. We commence the study in two main categories: first, the detection of micro-calcification; second, retrieval technology of calcification lesions in mammograms. The main research contents are as follows:(1)Detecting the calcification directing at he region of interest, in according the nature that micro-calcification are the high-frequency signals submerged in a very high-frequency noise and low-frequency background, we apply top-hat transform in the spatial domain to remove the most part of low-frequency background, then we use the open operation to remove the singular points, finally, we achieve a binary image by using threshold technology; In the frequency domain, we use multi-resolution analysis of wavelet transform and threshold technology and achieve a binary; At last, we calculate the two results by logic and operation, thus, we can eliminate low-frequency background and very high-frequency noise so that achieving the positioning of micro-calcifications. Experimental results show that the method get a higher positive rate and lower false positive.(2)We develop a new algorithm with multi-feature fusion and relevance feedback based on the study of single feature and feature fusion using single distance measure image retrieving techniques, this method adopts multi-distance measure to calculate the similarity directing at different features. Experimental results show that the method improved the retrieval performance effectively.(3)In order to integrate the algorithm to the system, the paper achieves the human-machine interaction function of the retrieval system by MATLAB GUI interface.