Research On Medical Image Mining Based On Improved Multi Kernel Support Vector Machine

Breast cancer is one kind of malign tumor that hazard female physically and mentally. At present, Image diagnosis among clinical diagnosis approaches seems to be the most suitable breast cancer diagnostic method in early stage. However, concealed information can not be discerned through doctor’s naked eyes. With the rapid expansion of Internet, online medical diagnosis has become one inevitability. Therefore, efficient data mining approach that is utilized to diagnose and distinguish medical images more precisely and quickly, promote classification accuracy, avoid misjudgment as much as possible, improve work efficiency for doctors has been deemed to be one important domain.This paper mainly studies muti-classification algorithms,Multi-Kernel Support Vector Machine,Directed Acyclic Graph Multi-Kernel Support Vector Machine and Node Selected Directed Acyclic Graph Multi-Kernel Support Vector Machine based on distributed scheme from data mining perspective. Meanwhile, such methods are utilized to construct crucial techniques and main algorithms in medical image field. This paper proposes a new method for weighted summation multi-kernel learning based on sample weighting, a new muti-class classification method based on DAG-WSMKSVM of nodes selection optimization and a new DAG-WSMKSVM of nodes selection optimization method based on MapReduce. Main works of this paper are as follows:(1) Method for weighted summation multi-kernel learning based on sample weighting is proposed. This paper proposes a new method for weighted summation multi-kernel learning based on sample weighting, which can attain weighting coefficients for each single kernel function by a regular sample learning method. Since kernel learning method in some complex situation can not caters for the requirements such as heterogeneous data or irregularity, large sample scale, sample distribution unevenness,etc, such proposed method could conquer those issues. On the other hand, it is a trend that we combine multiple kernel functions to expect to a better result. Each kernel function is weighted considering fit and adaptiveness of single kernel function for samples, which therefore generates summation multi-kernel decision function based on sample weighting. Classification experiments are respectively carried out on UCI machine learning standard data sets and mammography medical data sets through using the newly proposed algorithm, which achieves a relatively high accuracy compared with multi-classification method with traditional single kernel and multi-kernel.(2) A new multi-class classification method based on DAG-WSMKSVM of nodes selection optimization is proposed. A multi-class classification method based on DAG-WSMKSVM of nodes selection optimization is introduced in this paper. It selects the node by establishing alternative sets of nodes for every layer and chooses the nodes group which gets the highest training classification accuracy as the lower layer of current layer from the alternative sets of nodes, therefore the topology structure of DAG-SVM can be optimized and a better training and classification effect can be obtained. For an N-class classification issue, DAG-WSMKSVM can construct N*(N-1)/2 SVM classifiers( one classifier for a pair of classes). Concerning the situation DAG-SVM may behave poor due to the poor selection of nodes, so the new method based on node selection optimization(NSDAG-WSMKSVM) is proposed. Experimental results, comparing to other methods, show that the new method can achieve higher precision for UCI datasets and medical image multi-class classification.(3) DAG-WSMKSVM of nodes selection optimization method based on MapReduce is introduced. This paper also introduces DAG-WSMKSVM of nodes selection optimization method based on MapReduce. For defects in mammography classification through using NSDAG-WSMKSVM, efficiency issues on multi-user online synchronous diagnosis yielding mammography and timeliness improvement while confronting with large scale mammography, this paper ameliorates NSDAGWSMKSVM method quite effectively. A new classification approach called MRNSDAG-WSMKSVM based on MapReduce is constructed and put into practice on multi-class mammography classification, which achieves a desirable result.