Brain MRI Image Segmentation And Recognition

In recent years,medical imaging has developed rapidly,which mainly consists of ultrasound imaging(US),electronic computed tomography(CT),positron emission computed tomography(PET),magnetic resonance imaging(MRI)and so on.Because MRI can be imaged without injecting radioisotopes,it is safer than other imaging techniques such as CT and PET.With the development of imaging technology,computer-aided medical diagnostic technology has also begun to progress.At present,a large number of medical image processing algorithms have been proposed,which can help doctors make initial clinical diagnosis.The segmentation and recognition of brain MRI images are indispensable steps in medical image processing,so the research on the segmentation and recognition of brain MRI images is of great significance.At present,there are two main methods of brain MRI image segmentation.One is the traditional clustering technology and its improvement methods,whose main principle is to use different distance measurement formulas to classify pixels with similar features in the image into the same category;The other methods are based on the machine learning and deep learning algorithms widely studied in the computer field.These algorithms first need to manually or automatically extract some features of the images,and then perform multiple feature learning on the training images to obtain a certain rule to segment the images in the test set.These two types of methods only utilize the features of the image level and ignore the structural characteristics of the brain,so the accuracy of the segmentation has certain limitations.In recent years,machine learning algorithms and deep learning algorithms have been widely used in brain MRI image recognition.Traditional classifiers,such as support vector machines(SVM)and random forest classifiers(RF),need to select some features of brain MRI images artificially,such as gray level features,location features,texture features and so on.However,due to the variability of brain diseases,there are many handcrafted features that cannot represent image well.These characteristics not only can not help image recognition,but also may lead to recognition errors,which brings the risk of misdiagnosis.In addition,although deep learning algorithms can extract image features automatically that are beneficial to achieve high classification accuracy,training a deep convolution neural network requires a large number of training data sets.However,medical image data is very limited so that there is often over-fitting problems in the training process.In order to further improve the accuracy of brain MRI image segmentation and recognition,and to solve the problem that medical images is limited.This thesis has studied the two parts as follows:(1)This paper proposes a brain MRI image segmentation algorithm combining biometric,which is based on the prior knowledge of the H-shaped region of the central part of the cerebrospinal fluid(CSF).The CSF part is directly segmented using the Canny edge detection algorithm and the Fourier descriptors,and then the Fuzzy Local Information C-Means Clustering(FLICM)algorithm and the random forest algorithm are used to segment the gray matter and white matter.(2)This paper proposes a brain MRI image recognition algorithm that combines transfer learning and support vector machines.Firstly,the pre-trained convolutional neural network is used to extract the image features and then the SVM is used to classify the images.Image recognition does not need to train the entire deep convolutional neural network or artificially select image features,which solves the problem of overfitting caused by too few data sets.For the above two methods,we use the brain MRI image database published on the Internet to carry out experiments respectively.The experimental results show that compared with traditional methods and current advanced methods,the proposed brain MRI image segmentation and recognition algorithms have obtained higher accuracy.