Application Of Depth Convolution Belief Network In Classification Of Intracranial CT Images

Becouse of the rapid development of computer hardware and software technology,image processing,machine vision and pattern recognition technology axe becoming more and more diversified.With the continuous improvement of medical system information,medical data,the traditional shallow machine learning and pattern recognition algorithm has been unable to meet the damands of Process such a huge medical imaging samples.Deep Learning could learn the representation in large number of samples what makes deep learning model be a very useful tools in Internet information explosion era.With the rapid development of learning,Deep learning have been successfully applied in many fields.The combination of deep learning and computer aided diagnosis has become one of the main applications of computer vision and pattern recognition.The Conventional Deep Belief Network(CDBN)is an unsupervised deep learning model for image data which is developed by Restricted Boltzmann Machine(RBM).It combines the weight sharing of Convolutional Neural Network(CNN)and the characteristics of local visual field.At the same time,it adopts the unsupervised learning algorithm of Deep Belief Network(DBN),which has been obtained a good effect in image classification task.In this paper,We studied application of the convolutional deep belief network in the medical image classification task.The main contents are as follows:1.Appling CDBN to learning the representaton of intracranial hemorrhage CT images and classified the image using support vectors,which obtain a good performance.2.Proposed a parallel implementation of the CDBN in the CUDA framework.3.Using the Adaptive gradient algorithm to accelerate convergence rate of the CDBN while keeping a high accuracy of image classification.