Classification Of Brain CT Images Based On Deep Learning

With the rapid development of medical science,the further popularization of advanced medical equipment and the continuous innovation of medical imaging technology,the scale of medical image data is increasingly larger and shows a geometric growth.To store and maintain massive medical image data effectively has been a great problem,and it has brought a great challenge on precise retrieval of medical image data.At the same time,a lot of opportunities are created.Computer-aided diagnosis has become a research hotspot again.Traditionally the medical staff fix the medical image on a position to observe.However,the diagnosis result largely depends on the clinician’s clinical experience,and it is subjective,time-consuming and inefficient.In this paper,to improve the diagnostic accuracy and work efficiency of medical staff,the depth learning technology is used to construct convolution neural network and classify the brain images.Then the purpose of computer-aided diagnosis is achieved.The paper is organized as follows:1.Conduct an analysis on the structure of convolutional neural network in deep learning,and find out the factors that influence the efficiency and accuracy of the convolutional neural network in classification work.Based on the development framework of Caffe,a number of convolutional neural networks with different levels of structure are built.The build process includes the number of layers,the number and size of convolutional kernels,the size and mode of pooling,the choice of activation function and classifiers and the application of dropout layer.Then they are applied to the tasks of CIFAR-10 image classification.The results show that the more the network layers are,the number of convolutional kernels is larger;the smaller the convolutional size is,the the convolutional stride is shorter;the smaller the pooling size is,the pooling stride is shorter.The method to choose the maximum pooling is obtained.Leaky-ReLU function and dropout processing are used to improve the classification effect of convolutional neural network.2.According to the analysis of the structure of the convolutional neural network in deep learning,a 15-layer convolutional neural network model named CNN-BrainCT is built.It consists of seven convolutional layers,five pooling layers and three fully-connected layers.Part of the fully-connected layers have dropout layer.By changing the grayscale window ofthe brain CT image to simulate the three channels of the RGB image and the use of transfer learning,the CNN-BrainCT network is trained.Finally,this convolutional neural network is applied to the classification tasks of brain CT image with an overall accuracy of 67.64% and hit rate of 86.2% on the diagnosis of brain tumor disease.