Classification Of Brain Medical Image Based On Multi-feature Fusion

Nowadays, a growing number of brain diseases have affected people's life, and the research of brain diseases is becoming more and more important. Meanwhile, the rapid development of medical imaging technology and computer technology increase the huge demand for computer-aided diagnosis, in which the medical image recognition is one of the key technologies. Thus, the brain medical image recognition is one of the hotspots in the multi-disciplinary field of computer science and medicine.Since the inherent information which is extracted from the images directly affects the subsequent classification, the foundational work for image recognition is the phase of feature extraction. On the other hand, feature fusion is the trend of research in the medical image recognition. Therefore, this study mainly research on the feature extraction and feature fusion. Firstly, the thesis introduced the significance of brain medical image recognition and the current research worldwide; Secondly, the thesis introduced the basic knowledge of brain medical image recognition from three aspects: imaging technology, basic process of the image recognition and data sources; Thirdly, on the basis of the in-depth analysis of feature extraction technologies, a novel classification framework based on the multi-feature fusion was proposed; At last, we the support vector machine recursive feature elimination(SVM-RFE) algorithm was improved, and by introducing this improved algorithm to the proposed framework, we had achieved better classification results. Specifically, the main research works are as follows:1) This thesis summed up and evaluated the brain image recognition technologies including imaging technology and so on, and also analyzed the typical methods.2) On the basis of the comparative analysis of varieties of feature extraction technologies, a novel classification framework based on the combination of morphometric feature and texture feature was proposed to effectively improve the results of brain medical image recognition, and the adopted algorithms were gray level co-occurrence matrix(CLCM), Gabor and voxel-based morphometry(VBM).3) Multi-feature fusion technologies were further studied. Specifically, different feature reduction techniques, feature selection techniques and multinuclear learning methods were explored and compared. And an improved SVM-RFE algorithm was proposed to the classification framework, which makes us obtain the optimal feature subset to achieve better results.The results of comparative experiments on a public database have demonstrated the effectiveness of the proposed method—multi-feature fusion is better than the single feature, and the proposed feature selection algorithm could effectively extract the optimal feature subset to improve the result.