Research For Medical Image Classification Based On Association Rules

Diagnosis by medical image is one of the main methods of none wound diagnosis. Medical image has already been an important clinic diagnosis standard for disease diagnosis, decision-making before surgery and random visiting after surgery. Research on medical image diagnosis is a cross science between physic and computer fields and an important research direction. Data mining and computer technology are used to analyze, compute and process medical images; abundant feature information and rules are mined from medical images, which can aid doctors to diagnose and has upper academic value and broad application foreground.Currently, it is at the germ of mining medical image data and there are a lot of problems by applying existing mining methods directly to medical images. Researching and investigating some data mining methods and algorithms for medical data have some very important practical values.Destination, significance and abroad and domestic research status of medical image association rule mining fitting for medical images feature and its application are discussed and medical image association rule mining methods are analyzed. In order to overcome the problems of current association rule mining algorithms, we offer an improved mining algorithm based on classification tree and another improved data mining algorithm based on frequent pattern binary tree. Those two algorithms address the efficient problems of mining huge image data and build a basic classification system based on medical image features.This paper researches medical image data objects and those data mining from theory, algorithm and application three facts. Research fruits includes following contents:1 .It summarizes medical image feature extraction completely, analyzes and extracts wavelet features, gray co-occurrence matrix features, histogram features and density function features according to the features of medical images for medical image classifier.2. In order to overcome disadvantages of Apriori based algorithms that need to access database many times and produce a lot of candidate item-sets, a classification tree based algorithm is designed. It can deduce the number of mining data and improve mining efficiency. At the same time, it takes less main memory cost by classification tree.3. In order to overcome disadvantages of FP-GROWTH based algorithms that take too much main memory, a binary frequent pattern is designed. It takes less main memory cost and is fit for association rule mining great number of data like image data.4. Using algorithms of Apriori, FP-GROWTH and their improved algorithms, we do some association rule mining on medical image data and make use of rules of high interesting degree to build a basic medical image classifiers.