Medical Images Denoising Method Based On Deep Convolution Neural Network

Medical image-assisted diagnosis plays a very important role in the process of modern medical diagnosis.The radiologist uses non-invasive medical imaging methods to obtain images of some tissues in the patient’s body,which is not only convenient to obtain,but also can minimize the harm to patients during the diagnosis process.Existing medical images are mainly divided into magnetic resonance imaging(MRI),X-ray images,and computed tomography(CT).However,in the process of acquiring medical images,not only will it be affected by various types and degrees of noise,but there will also be stripe artifacts,which will interfere with the diagnosis,analysis and treatment of diseases to a certain extent by doctors.Therefore,medical images denoising and improving the accuracy of denoising as much as possible have very important scientific research significance and clinical use value.This paper mainly studies the application of deep convolutional neural network(CNN)in medical images denoising.The main research work and results are as follows:(1)We summarized the research status of medical images denoising at home and abroad and several mainstream denoising algorithms are analyzed in depth.The traditional denoising algorithms include non-local mean(NLM),non-local full variational regularization algorithm(NLTV),and block-matching and 3D filtering(BM3D).Denoising algorithms based on machine learnin,which include noise reduction autoencoder(DAE),multi-layer perceptron image denoising model(MLP),and method(Dn CNN)using residual learning and batch normalization(BN).The advantages and disadvantages of these denoising methods are further analyzed.(2)Focus on the research of deep convolutional neural network model and analyze its advantages in medical images denoising.We further optimized the existing deep convolutional neural network by the BN,residual learning,multi-scale parallel extraction of noise and other methods.The X-Re CNN model is proposed for X-ray image denoising and the CT-Re CNN model is for CT image denoising.Experimental results show that the X-Re CNN and CT-Re CNN are better than other algorithms mentioned above in both subjective visual effects and objective evaluation indicators.Not only can the noise and artifacts in the medical image be effectively removed,but the structural details in the medical image can be better preserved.(3)In order to further reduce the model complexity and shorten the calculation time for model training and test,the paper also uses depthwise separable convolution and dilated convolution to lighten the model.The experimental results show that the lightweight network model can not only complete the denoising task excellently,but also shorten the calculation time.