Research On Key Technology Of Classification For Medical Image

With the development of medical imaging technology, many hospitals widely useimaging equipment which makes lots of medical image data and relative medical parametersaccumulated. We can use data mining technology to dig out the useful information from hugemedical image data. The useful information not only assist physicians to make effectivemedical diagnosis but also make medical experience and knowledge shared to all people vianetwork.Firstly, this paper analyzes the current research on medical image mining. Secondly, itfocuses on the relevant technology based on medical image mining. At last, it starts theresearch on key technology of classification for CT image of human brain.The research work and innovations of this paper can be organized in the follows aspects：(1) Medical image feature expression. Incorporating domain knowledge of medical field,we extract the ROI in medical image and the feature attributes of ROI, then we group theROIs to clusters by cluster method. At last, the medical images are organized by medicalimage feature database.(2) It researches on two ensemble classifiers which are widely used at current. One is theAdaBoost algorithm which has serial structure and the other is Random Forest algorithmwhich has paral1el structure.(3) As relative costs from different types of misclassification error are inherently unequalin medical image classification, it proposes the cost-sensitive ensemble classificationalgorithm which is an improved algorithm for the AdaBoost algorithm. It uses the proposedalgorithm to make classification for medical image.(4) It proposes an improved algorithm based on random forest algorithm forclassification. This paper leverages key elements of the derivation of generalization errorbound to derive bounds on detections rate(DET) and false alarms rate(FAR) on ROC andgives the performance optimization guidelines for tuning class-specific correlation inferredfrom the bounds for each region. At last it uses the algorithm for medical image classification.