Multi-mode Mri Brain Tumor Segmentation Based On The Feature Fusion

In medical imaging, magnetic resonance imaging(MRI) is an important imaging technology, the imaging technique has a high quality image display effect and has been widely used in medical diagnosis of body tissues and organs lesions, especially the detection of brain lesions. MRI brain tumor segmentation provides a great help in the clinical diagnosis, it is a difficult problem for brain tumor to be segmented out of brain tumor more quickly and accurately. Because of the variability and complexity of the brain tumor magnetic resonance images, and the size and shape of brain tumors, it is important to extract brain tumor feature. In recent years, there are also many studies on this aspect. Although it has good results, it also has field and wide space to promote.This paper mainly studies the MRI brain tumor segmentation algorithm:The extraction of brain tumor from Gabor wavelet and CNN use feature fusion method to form a new feature. Then it introduces a new dimension reduction methods to reduce feature based on the improved algorithm and has achieved goods results. The specific work content as follows:( 1) Analysis the basic principles of Gabor wavelet algorithm, using 40 Gabor filter with 5 scales and 8 orientations of convolution on brain slice image to extract feature, and make the convolution results as feature vector. Then it verifies brain tumor segmentation based on SVM and the segmentation results do post-processing. Then use the method into the GBM data and analysis its result.( 2) Analysis the basic principles of convolution neural network, introduce the network structure and training process of convolution neural network. The convolution layer can enhance the original signal strength and reduce the noise because of convolution operation. In the down sampling layer has a sampling process. This method premise not reduce the useful information and reduce the amount of data need to be processes. In the convolution neural network also reduce parameters and share weight that can improve the speed of operation. Then use the method into the GBM data and analysis its advantages and disadvantages.( 3) In order to improved the accuracy of segmentation, the paper puts forward the method that combine the features from artificial selection and machine learning. Expand the relevant knowledge of feature fusion, For each tumor sample, using serial combination of feature fusion in series into a column vector as feature vector from the feature make from Gabor wavelet and CNN, at the same time based on kernel entropy component analysis to reduce the dimension of the new features after fusion, and then classified the features before and after dimensionality reduction by SVM. Analysis its advantages and disadvantages according to the experimental results.