Brain Tumor Segmentation Using Convolutional Neural Networks In MRI Images

Gliomas are the most common type of brain tumor and the most aggressive.High-grade gliomas can also cause high mortality.Therefore,the treatment plan for tumor is the most important to improve the quality of life of tumor patients.Advanced tumors are the most frequent primary brain cancers.Each year,85% of new cases are diagnosed as malignant primary tumors.MRI plays an important role in the detection and treatment of brain tumors.In particular,accurate delineation of the boundaries of the tumor’s visible exudate can be useful in quantifying the size of the lesion and its cumulative rate of longitudinal data.Record the inhomogeneous complex,However,noting the anisotropic complex shape that high-grade gliomas can take due to their high diffusion and proliferation rates,as well as their appearance variability,the segmentation of the complete glioma including all its compartments becomes more challenging.Nuclear magnetic resonance imaging is an extremely effective technique for evaluating the application of tumors.The use of neuro-imaging techniques to diagnose brain tumors and to detect the boundaries between visible and invisible tumor cell infiltrates has motivated the emergence of different tumor segmentation algorithms.Automated segmentation becomes difficult by recording strong variability in tumor appearance and shape.In this article,in order to split the lesion of the tumor,we separate the two convolution neural network and support vector machine(SVM)classification model,and using these two models for segmentation,more specifically,we put forward a kind of migration based on convolutional neural network learning to the method of support vector machine classifier capture them complete characteristics at the same time.Our framework consists of two phases in series.In the first stage,we trained the convolutional neural network to learn the images from the image space to the tumor marker space.In the test phase,we used the expected mark output from the convolutional neural network,and sent the grayscale image along with the test to a support vector machine classifier for accurate segmentation.Then we made our deep convolutional neural network-support vector machine serial classifier for iteration.In our experiments,we use the same training data set of convolution neural network model(CNN),support vector machine(SVM)model and deep convolution neural network-support vector machine(CNN-SVM)model of the three kinds of model training,the training is completed,we used the same set of test data of three kinds of model test,finally got three sets of different data.The experimental data show that the proposed deep convolutional neural network-support vector machine model has better segmentation effect than the support vector machine and convolutional neural network models.