Research On Medical Image Denoising Algorithm Based On Deep Convolutional Neural Network

In the process of medical image acquisition and transmission,they are often affected by the interference of imaging equipment and external environmental noise,which reduces the quality of the images.These images affected by noise,which are of poor quality,have a great impact on the analyzation of medical images and clinical diagnosis.Therefore,medical image denoising is one of the most important tasks in the pretreatment process of medical images.In the field of computer vision technology,with the fast development of deep convolutional neural networks,the medical image denoising algorithm based on the neural network has drawn more and more attention.One of the obstacles to applying deep convolutional neural networks to the medical image denoising is the tremendous need for training samples that are not affected by noises.However,many medical images have no original images.Recently,Ulyanov came up with an algorithm,Deep Image Prior(DIP),which does not need prior training before recovering the images.The algorithm shows that convolutional neural networks have the ability to capture low dimensional image information of a single image,but this algorithm is not ideal for medical image denoising.It is a huge challenge to denoise from medical images quickly without a large number of training samples.In this paper,the optical coherence tomography(OCT)image will be taken as the start point,with the main content be divided into these aspects as followed:(1)A medical image denoising algorithm based on non-local deep convolutional neural networks will be proposed to tackle the problem that DIP could not denoise efficiently for medical images.Ordered non-local variables will be added to the DIP network study as the regularization loss term to maintain more information about image structure,so that clearer denoise images can be obtained in the network training process.(2)In consideration of the similarity between image frames,a three-dimensional non-local deep convolutional neural network medical image denoising algorithm will be proposed as the expansion of the medical image denoising algorithm based on non-local deep convolutional neural networks.This algorithm could learn the non-local similarity of multi-frame OCT images during deep convolutional neural network training,and achieve a more ideal denoising effect.(3)Although the medical image denoising algorithm based on non-local deep convolutional neural network and the three-dimensional non-local deep convolutional neural network medical image denoising algorithm can solve the problem that the convolutional neural network requires a large number of training samples,it has also achieved better medical images denoising effect.But they are difficult to meet the needs of real-time processing in practical applications.In order to be able to denoise medical images in real time,this paper first studies the basic principle of Deep Convolutional Neural Networks(Dn CNN).Secondly,the results of three-dimensional non-local deep convolutional neural network medical image denoising algorithm will be taken as training samples,combined with multi-scale feature extraction method to improve it,to have more feature information be extracted from the image.Experiments show that the multi-scale feature extraction deep convolutional neural network can not only be applied to medical image denoising,but also has great advantages in computing speed.