Research On Techniques Of Multi-angle Medical Image Retrieval

Multi-View medical imaging is a medical image taken from different angles.The main purpose of the retrieval of medical images is to help doctors diagnose.The traditional image retrieval method uses single image, but it’s no longer ableto achieve the requirements of the people with the increasing and updating ofmedical imaging case data. At the same time, the image information of singleimage has limitations and the performance of retrieval is affected. Therefore theretrieval of medical image using multi-angle, combined with the imageinformation of the plurality of images can be more effective.Breast disease is the most common tumor diseases threaten women’s health.Early diagnosis and detection become main measures for the treatment of breastcancer effect optimum. The importance of early detection and diagnosis basedon the X-ray image of the breast, masses and calcifications become signs ofbreast cancer. A breast X-ray image have axial (referred to as CC bits) view andoblique (referred to MLO bit) view in breast medical image. MLO view imagescan display clearly breast lesions details that can’t display in CC view, and canalso be a good show even if the lesions occur in the chest. This thesis mainlyresearch for the breast cases of mass, the suspicious lumps regional that in thedifferent angles breast view pairing achieve mammography multi perspectiveretrieve.Firstly, match two-view images of the breast cases mass for multi-viewbreast image retrieval. The theoretical basis is the same as the in different anglesbreast image match or mass lesion area in the central axis of the projection pointto the distance between the nipple remained basically unchanged. Here the axisrefers to the flight to a vertical segment of the chest wall line of the shortestdistance from the nipple in the breast area. Selecting the nipple and chest wallline as a reference position to establish the matching region, the matching band using layered extracted suspected mass region. Using weighted mutualinformation to achieve tumor of the breast image matching. Secondly, amulti-view breast image retrieval system is built. This system contains theestablishment of the query image database, inquiring image sample set, retrievalmode and the similarity measure selection. The query image database contains200images and the query image sample set consists of20images. Retrievalmode includes SCC, SMLO, MCR and MAR four kinds. Thesis use Euclideandistance, Cosine distance, Correlation distance, K-nearest neighbor andK-nearest neighbor combined with mutual information of five kinds distancemeasure in the image library, and compared five retrieval methods of retrievalperformance. Finally, the thesis uses the search interface of Microsoft VisualStudio2010to design human-computer interaction, and Matlab programming toachieve multi-angle breast image retrieval.In order to prove the effectiveness of the multi-view the breast imageretrieval method, experiments have been conducted to verify each part of thethesis. Experimental results show that the method is a valuable and effectivemethod of computer-aided diagnosis.