Research And Application Of Medical Image Processing Based On Deep Learning

Medical image segmentation is the most complex and critical technology in medical image processing.It has the important clinical practical value and research significance.In recent years,with the rapid development of deep learning technology,convolutional neural network has become the mainstream image segmentation method.Network models such as fully convolutional neural network(FCN)and U-Net have been proposed successively.The U-Net network is a convolutional neural network with good performance,but when extracting features from medical images,due to continuous downsampling and upsampling,the problem of image detail features disappearing easily occurs.In response to this problem,this paper proposes an improved image segmentation algorithm of U-Net network,which uses multi-scale and attention mechanism to replace the convolution blocks and skip connections in the original network to improve the accuracy of image segmentation.In addition,considering that medical images will be polluted by environmental noise during imaging or acquisition,this paper proposes an image denoising method for medical image preprocessing.The primary substance of this study is as hereunder mentioned:(1)Aiming at the problem of low signal-to-noise ratio and loss of edge image details in medical images,this paper proposes a residual dense block-based convolutional neural network(DRCNN)image denoising algorithm.The model removes the noise in the image by introducing multi-level residual network and dense connection,and also uses Leaky Re LU activation function for the overall network,while better retaining the effective information of the image and effectively avoiding feature loss.Compared proposed model with the deep convolutional neural network residual learning(Dn CNN)model,the proposed algorithm in this paper shows that the DRCNN algorithm has an average increase of about 3.7d B in the peak signal-tonoise ratio on the medical image Lung testsets,and an average increase of about 0.0175 in the structural similarity.Experiments show that the DRCNN algorithm performs better in the denoising effect of medical image datasets.(2)Aiming at the problem of U-Net losing some shallow image spatial features,this paper proposes a medical image segmentation model IRS-Net based on multi-scale feature attention mechanism encoderdecoder.The multi-scale strategy and attention mechanism are introduced into the U-Net network model,and an efficient multi-scale feature extraction block is designed to replace the encoder and decoder.The multi-scale feature extraction block combines the high-level semantic information and low-level semantic information in the image.The information is fused to enrich the contextual information.The skip connection part between the encoder and the decoder is replaced by an attention mechanism module to extract more spatial feature information of the image.The experimental results show that the IRS-Net algorithm can accurately segment organs in medical images,and the algorithm’s F1 score,Accuracy,Specificity,Precision,ROC and AUC are all better than FCN,U-Net,SA-UNet and DRNet image segmentation models.(3)Design a medical image segmentation aided diagnosis system.Based on the above research results of medical image denoising and image segmentation model,this paper designs and implements a medical image segmentation auxiliary diagnosis system.The system mainly includes two main modules:medical image denoising and medical image segmentation.By building an auxiliary diagnosis system platform,reducing misdiagnosis and missed diagnosis,and providing auxiliary work for doctors’ diagnosis.