Research On Medical Image Projects Association Model Mining And Retrieval Technology

As a cross research field of computer science and medicine, retrieval ofmedical images and diagnostic technology research become an importantresearch direction in the field of computer science as well as medical at homeand abroad. Internal features and rules in images could be excavated through datamining techniques, so as to assist doctors to do similar case comparison anddiagnosis, which has high academic value and wide range of applications. Atpresent, a few research on medical image using data mining technology, andthere are many problems when using the existing data mining techniques to doresearchs on medical image. It has a realistic and important significance toprocess medical image with data mining methods efficiently and accurately.Effective image segmentation could guarantee the credibility of theextractive image features, this thesis proposed an adaptive region growingmethods, using the association rules to extract image features, and improved theassociated classification engine (ACE) algorithms to achieve the classification ofmedical images.The main contents are as follows. Firstly, an adaptive region growingmethod for mass segmentation is presented so as to solve the problem of massimage segmentation. Fit out the background of the mass ROI area. And then,subtract the ROI and its background to get the ROI after the pretreatment. Finishmass area segmentation through the adaptive region growing method which isbased on the gradient of images. Secondly,32shape and texture features areextracted from each ROI,18features are selected through feature selectionalgorithm based on association rules, which performs better than other featureselection algorithm in retrieval. Thirdly, establish and achieve a medical imageassociated classification model. Discretize continuous features of images throughthe method of minimum description length (MDL) to extraction classification association rules. The medical image associative classification has been realizedthrough the improved ACE algorithm, which improved the classificationaccuracy.At last, a region of interest (ROI) database containing219ROI and acorresponding feature database are presented in the dissertation, which provides aconvenient ROI query and update system to users.