Research on CT Image Super-resolution Reconstruction with Multi-branch Fusion Network

Low-dose CT scan is an effective way to reduce the radiation risk of patients, but it can reduce the quality of images and affect the accuracy of medical diagnosis. To improve the resolution of CT images, we proposed a multi-branch fusion network to reconstruct CT images. First, the whole network framework is constructed by cascading multiple sub-networks. Local residual learning is introduced into each sub-network to improve the accuracy and convergence speed. Second, three branches of the convolution operation are used to extract the feature information of the image in parallel, the information is fused as the input of the next sub-network, and high-resolution images are reconstructed finally. By extending the depth and width of the network, the extracting feature ability of the network is improved. Meanwhile, global residual learning is introduced to enhance the reconstruction ability of the network. We used the classical LUNA16 datasets to train the model, calculated the loss function between the target image and the ground truth image, updated the weight, and got the optimized network model. We compared the model with related algorithms, the results showed that our method achieves better reconstruction effect, which will improve the accuracy of medical diagnosis.