Data Mining On Medical Images

Multimedia data mining is an important branch of data mining. This dissertation focuses on main techniques and algorithms of data mining based on medical images. On mammogram database, the contrast enliancement of mammogram and evaluation measures, feature extraction and selection and classification algorithms are explored. The research work can be organized in the following aspects.1. Regarding the medical images, the transformation, feature extraction and reducts of image data are studied by integrating digital image processing and pattern recognition. The prototype and mechanism of medical image data mining is presented under the support vector machine and rough set theory.2. A contrast enhancement on uniformity is proposed especially for dense breast tissue, aiming to make suspicious area more visible. This new technique is compared with two other existing contrast enhancement techniques—morphological enhancement and histogram equalization—and is found to perform better than either.3. A set of features containing almost all information of breast mass is computed for mass detection and classification, which are the precondition and element of computer-aided diagnosis.4. PC A (Principal Component Analysis) is an important method for feature projection and pattern dimension reduction in statistical analysis. And rough set approach has the capability in reducing significantly attributes. Our algorithm is based on an application of a rough set method to the result of K-L transformation. Experiment result shows that the features selected by proposed algorithm can be a help in the improvement of classification.5. Proximal support vector machine (PSVM) for medical image data mining is put forward, which not only runs faster than standard support vector machine classifiers but also is easy to implement with satisfying results.6. A novel algorithm of SVM, DFP-PSVM, for unbalanced two-class dataset is presented in this dissertation by transforming restrained mathematics programming into unrestrained mathematics programming. The method solves the influence that the class with much more set of data prevails over the class with much smaller set of data in 2-category classification.This research is supported by the National Natural Science Foundation of China.