Intelligent Classification Algorithms On Medical Images

Data mining can be used to process medical images in order to collect models, build relations, rules, and find changes from mass amount of data, which would also reduce the duration of diagnosis and improve the accuracy of diagnosis made by doctors. This dissertation focuses on main techniques and algorithms of applying data mining to processing medical images. Based on mammograph database, the preprocessing, feature extraction and selection, as well as classification algorithms are explored.The research work can be organized in the following aspects.1. Using with techniques of data mining and digital image processing, this dissertation examines the transformation, feature extraction, reduction and pattern recognition of image data, and proposes a prototype and mechanism of how to apply data mining techniques to process medical images.2. A set of feature vectors containing almost all information of mammograph are established. Based on these feature vectors, scalar data can be discretized and feature detection can be done using fuzzy clustering algorithm, which is the precondition of computer-aided diagnosis.3. Based on decision tree algorithm, the concept of Priority of Attributes is presented. Using this concept, mammograph can be categorized and typical image data of breast cancer can be recognized. In this way, the accuracy of categorizing could be greatly improved.4. The algorithm of association rule and association rule classification are thoroughly examined. Regarding to the large scale of calculation using Apriori Algorithm, an approach of adding restriction to the attributes of extended items is presented, which can be used to categorize mammograph.5. Rough set approach can be used for feature projection and pattern dimension reduction based on discernibility matrix and function. In this dissertation, Rough set approach combined with association rule algorithm are proposed in association rule abstraction. In addition, a new binary algorithm based on particle to categorize mammograph is presented. The result of experiment shows that the proposed approach can greatly improve the accuracy of classification.This research is funded by the National Natural Science Foundation of China.