Super-resolution Reconstruction Of Medical Image Based On Deep Learning

High resolution(HR)medical images can provide detailed structure information of organs or tissues,which help clinical diagnosis,decision-making and accurate quantitative image analysis.Due to the limitation of physical conditions such as hardware equipment,medical imaging has disadvantages such as long scanning time and low signal-to-noise ratio.Therefore,the low resolution(LR)images produced by the imaging equipment are converted into high resolution images by means of algorithm,that is,the super resolution(SR)reconstruction algorithm of medical images can solve the above problems more effectively.Deep learning is faster and more accurate than traditional algorithms,so this paper uses the super-resolution reconstruction technology based on deep learning.By analyzing the principle of convolutional neural network and comparing the different reconstruction algorithm based on convolution neural network research,this paper selects generative adversarial network(GAN)as the basic network framework and employ residual dense block(RDB)as the basic block of the generator to solve the problems existing in the reconstruction algorithm,such as the network model is simple and the texture details are not clear enough.This network combines the advantages of dense network and residual network to increase network depth and make better use of the feature information of each layer.The network loss function is the weighted sum of content loss and confrontation loss so that the network can generate super resolution images with fine texture.Anatomical marker segmentation and pathologic localization are important steps in the automatic analysis of medical images,and the generation of high-quality super-resolution medical images is the precondition to realize these steps.This paper not only compares the reconstruction results of medical images generated by different algorithms,but also carries out the contrast experimental results of retinal blood vessel segmentation in retinal images reconstructed by different algorithms.The accuracy of the segmentation results can also reflect the reconstruction effect.Finally,the experimental results of reconstruction of retinal image and cardiac MRI show that the super-resolution images generated by this algorithm has finertexture details and is closer to the original high-resolution image in the objective evaluation index and visual perception.