Research On Association Pattern Mining Algorithm For Medical Image

With the increase of medical image data, people expect to find the useful information which is hidden within immense amounts of image data. This useful information can help physician to make the diagnosis well or it will be convenient to research on certain illness for experts.First in this paper, it analyzes the current domestic and foreign research on medical image mining. Second, it studies on the key technologies and relevant algorithms which aimed at medical image mining. Last, it finds the association pattern for the region of interest (ROI) in CT image of human brain.The research work and innovations of this paper are as follows:(1) Medical image pre-processing. First, it extracts the ROI in medical image using the threshold method. Second, it defines the feature attributes of ROI in accordance with the domain knowledge of medical image and extracts these feature attributes using Matlab language. Last, it groups the ROI to clusters by clustering method.(2) It summarizes the current development status of association rules mining techniques and researches on the two classical methods. One is the Apriori algorithm which generates frequent itemsets from candidate itemsets and the other is FP-growth algorithm which generates frequent itemsets directly.(3) It proposes the Diff_FIMA algorithm which bases on the difference constraint method. Certainly, Diff_FIMA is an improved algorithm for the FP-growth algorithm and can find association patterns for ROI in CT image of human brain.(4) It proposes an improved algorithm GMA (association graph and feature matrix algorithm) which is an Apriori-like algorithm. GMA algorithm improves the efficiency in generating candidate itemsets and computational complexity. Finally, it uses GMA algorithm to mine association patterns for the ROI in CT image of human brain.