Abstract: Improving the sharpness of medical images is of great significance for doctors to quickly diagnose and analyze the disease. A medical image super-resolution reconstruction algorithm based on residual network is proposed to fully improve the texture details of medical images. Firstly, this paper selects the appropriate data set, uses the very deep convolution neural network, cascades several smaller filters, extracts the information from the image adequately. Secondly, the residual learning method and the Adam optimization method are used to accelerate the convergence of the deep network model. Finally, training sets of different magnifications are combined into a hybrid data set for training, which improves performance while greatly reducing the number of parameters and training time. The experimental results show that the PSNR, SSIM and FSIM of the proposed algorithm are higher than the existing algorithms, the reconstructed image has more abundant details and more complete edges.

Key words: super-resolution, deep learning, medical image, residual network

摘要： 提高医学图像的清晰度对于医生迅速的做出病情的诊断与分析具有重要的意义，为充分提高医学图像的纹理细节清晰度，提出一种基于残差网络的医学图像超分辨率重建算法。选取合适的数据集，使用非常深的卷积神经网络，多次级联较小的滤波器，充分提取图像中的信息；使用残差学习的方式以及Adam优化方法来加快深层网络模型的收敛；将不同放大倍数的训练集组合成混合数据集进行训练，提高性能的同时大大减少了参数数量与训练时间。实验结果表明，所提算法的PSNR、SSIM、FSIM均高于现有的几种算法，重建出的图像细节更加丰富，边缘更加完整。

关键词: 超分辨率, 深度学习, 医学图像, 残差网络