Application Of Model Based On Convolutional Neural Network In Medical Images

With the development of artificial intelligence,deep learning as its very important branch has made a qualitative leap in automatic speech recognition(ASR),natural language processing,and image processing in recent years.Deep Belief Network(DBN),Convolutional Neural Networks(CNN),and Convolutional Restricted Boltzmann Machine(CRBM)based on a convolutional neural network have been widely used as a classic model for deep neural network.There is a popular research topic about medical images processing.The common pre-criteria measure for breast-related diseases is Breast X-ray image analysis,which uses the deep-learning algorithm to calculate and process mammographic X-ray images,to analyze the details of the images that are difficult to distinguish with the naked eye,and further to improve the accuracy of image classification.Firstly,the main work of this thesis is to study the existing CNN model and compare the algorithms of some neural network layers such as single-core convolution,multi-channel convolution,and multi-scale convolution in convolution layer,and max-pooling,average-pooling,spatial pyramid-pooling in the pooling layer.Secondly,after studying and comparing these learning algorithms,a multi-scale hybrid pooling convolutional neural network(MSHP-CNN)is proposed for the problem that the CNN model is not robust under different image samples.Meanwhile,on the basis of learning the CRBM model,a parallel optimization algorithm for CRBM model based on Spark framework is proposed for the problem of long training time.The main research contents of this article are as follows:(1)Proposing a MSHP-CNN algorithm based on convolutional neural network model.The classic CNN model has achieved a great breakthrough in the calculation and processing of graphic images,but its low classification accuracy still exists.In order to improve the classification accuracy,the neural network model training based on MSHP algorithm of CNN model is proposed.Through the experiments on MNIST data set and CIFAR-10 data set and by comparison with the standard CNN model learning algorithm and the improved MS-CNN,the result of this thesis shows that the image classification accuracy of the analogy(Multi-Scale CNN)algorithm improves by 2%.(2)Proposing a parallel optimization algorithm for CRBM model based on Spark framework.The long training time of CRBM model is the main problem of the model.In order to reduce the training time and improve the training efficiency,the CRBM model parallel optimization algorithm based on the Spark framework is proposed in this thesis and it is performed on the MIAS mammography data public data set.The result shows that the parallel optimization algorithm is less than the original one-third of the training time compared with the non-optimized CRBM algorithm in the same classification accuracy.