Classification Of Medical Images Mode Based On Deep Learning

The rapid development of modern medical science and technology to make medical images in clinical diagnosis plays an increasingly important role, while the medical image mode is also more diverse, and the overall number of medical images are presented every day exponential growth, this aspect of the disease to bring diagnostic medical imagery reliable data, but on the other hand how to manage these image data are stored and how to better explore the use of these image data value in itself has become an urgent need to address the problem. Therefore, the medical image pattern classification becomes effective implementation of modern medical image storage management, one important way to explore the value, but the traditional manual classification can not meet the growing demand for medical images, the urgent need for the use of modern information technology and computer technology, automatic classification of medical images management, excavation and other useful information to improve management efficiency and utilization value of its own. Therefore, it is possible to achieve effective management of medical image classification algorithm has become a hot research.This paper reviews the present situation of the development of medical images and medical image classification of diseases, were investigated after extraction method based on the traditional characteristics of medical image classification algorithm based on the depth and the depth of learning belief networks, automatic stack machine and convolution coding nerve net medical image classification algorithm, and comparative analysis of the classification results. Finally convolution belief networks to overcome the above-mentioned network structure is insufficient, and through specific experiments demonstrate the effectiveness of the algorithm. The main contribution of this paper are:(1) using image processing and analysis techniques for medical image analysis, and extract a variety of different characteristics of medical images, comparative analysis of its application in medical image classification. Specifically, for medical image texture feature extraction mainly local binary pattern feature and use GLCM method of extracting features of different matrices property value. For the main local features extracted SIFT features, and the use of traditional artificial neural network classification algorithm, classification accuracy compare different features.(2) research and analysis of medical image classification shallow network structure model based on the first feature shallow learning network and Restricted Boltzmann Machine, automatic coding machine shallow sparse network structure model is introduced, after which X-ray images of breast cancer database MIAS experimental data, the database area automatic focus feature extraction shallow respectively through different networks, get a different kind of image features, and then take advantage of these features and characteristics of the image processing methods enter the combined neural network classifier, comparative analysis of the experimental results.(3) analysis of the medical image classification method based on the depth of learning, the depth of learning three network structure model-the depth of belief networks, Stacker automatic coding machine, convolution neural network analysis reports, and proposed based on deep learning X-ray images of the breast tumor classification method to further improve the accuracy of medical image classification, experimental results show that, compared with the traditional method based on neural network classification feature extraction, classification performance better.(4) The depth of the network structure learning mode for scalable performance classification of medical images, and for the deep belief networks for medical image classification is not sensitive enough detail the characteristics of the input image size is limited, convolution confidence proposed network of medical images the method of classification and LIDC/IDRI lung image database is verified, it is proved the validity of the method.Through different research work of this article, show that shallow network model can improve the accuracy of medical image classification to a certain extent; the depth belief networks, Stacker automatic coding machine, convolutional neural network learning network as the representative of the depth of the structural model not only can effectively avoid the traditional medical image manually design features tedious steps, and classification accuracy has been significantly improved. The convolution operation and depth of belief networks combined belief networks generated convolution depth study to further improve the network structure model in the medical image.