The clarity of medical images directly affects the clinical diagnosis. Due to the limitations of imaging equipment and environmental factors， it is often impossible to directly obtain high-resolution images， and the hardware of most smart terminals is not suitable for running large-scale deep neural network models. Therefore， a lightweight dense neural network model with fewer layers and parameters was proposed. First of all， dense block and skip layer structure were used in the network for global and local image feature learning， and more feature information was introduced into the activation function， so that the shallow low-level image features in the network were able to be propagated to the high-layers more easily， thereby improving the super-resolution reconstruction quality of medical images. Then， the multi-stage method was adopted to train the network and the dual-task loss was used to strengthen the supervision and guidance in network learning， which solved the problem of difficulty increase in network training caused by highly magnified image super-resolution reconstruction. Compared with Nearest Neighbor （NN）， bilinear interpolation， bicubic interpolation， Convolutional Neural Network （CNN） based algorithm and the residual neural network based algorithm， the proposed model is of high practical value on better reconstructing the texture details of medical images， achieving higher Peak Signal-to-Noise Ratio （PSNR） and Structural SIMilarity （SSIM）， as well as achieving good result in both training speed and hardware consumption.

Aiming at the problems of low accuracy and poor robustness of traditional point cloud registration algorithms and the inability of accurate radiotherapy for cancer patients before and after radiotherapy， an Attention Dynamic Graph Convolutional Neural Network Lucas-Kanade （ADGCNNLK） was proposed. Firstly， residual attention mechanism was added to Dynamic Graph Convolutional Neural Network （DGCNN） to effectively utilize spatial information of point cloud and reduce information loss. Then， the DGCNN added with residual attention mechanism was used to extract point cloud features， this process was not only able to capture the local geometric features of the point cloud while maintaining the invariance of the point cloud replacement， but also able to semantically aggregate the information， thereby improving the registration efficiency. Finally， the extracted feature points were mapped to a high-dimensional space， and the classic image iterative registration algorithm LK （Lucas-Kanade） was used for registration of the nodes. Experimental results show that compared with Iterative Closest Point （ICP）， Globally optimal ICP （Go-ICP） and PointNetLK， the proposed algorithm has the best registration effect with or without noise. Among them， in the case without noise， compared with PointNetLK， the proposed algorithm has the rotation mean squared error reduced by 74.61%， and the translation mean squared error reduced by 47.50%； in the case with noise， compared with PointNetLK， the proposed algorithm has the rotation mean squared error reduced by 73.13%， and the translational mean squared error reduced by 44.18%， indicating that the proposed algorithm is more robust than PointNetLK. And the proposed algorithm is applied to the registration of human point cloud models of cancer patients before and after radiotherapy， assisting doctors in treatment， and realizing precise radiotherapy.

To solve the problems of image detail loss and unclear texture caused by interference factors such as noise， imaging technology and imaging principles in the medical Magnetic Resonance Imaging （MRI） process， a multi-receptive field generative adversarial network for medical MRI image super-resolution reconstruction was proposed. First， the multi-receptive field feature extraction block was used to obtain the global feature information of the image under different receptive fields. In order to avoid the loss of detailed texture due to too small or too large receptive fields， each set of features was divided into two groups， and one of which was used to feedback global feature information under different scales of receptive fields， and the other group was used to enrich the local detailed texture information of the next set of features； then， the multi-receptive field feature extraction block was used to construct feature fusion group， and spatial attention module was added to each feature fusion group to adequately obtain the spatial feature information of the image， reducing the loss of shallow and local features in the network， and achieving a more realistic degree in the details of the image. Secondly， the gradient map of the low-resolution image was converted into the gradient map of the high-resolution image to assist the reconstruction of the super-resolution image. Finally， the restored gradient map was integrated into the super-resolution branch to provide structural prior information for super-resolution reconstruction， which was helpful to generate high quality super-resolution images. The experimental results show that compared with the Structure-Preserving Super-Resolution with gradient guidance （SPSR） algorithm， the proposed algorithm improves the Peak Signal-to-Noise Ratio （PSNR） by 4.8%， 2.7% and 3.5% at ×2， ×3 and ×4 scales， respectively， and the reconstructed medical MRI images have richer texture details and more realistic visual effects.

Computed Tomography （CT） three-dimensional reconstruction technique improves the quality of three-dimensional model by upsampling volume data， and reduces the jagged edges， streak artifacts and discontinuous surface in the model， so as to improve the accuracy of disease diagnosis in clinical medicine. A CT three-dimensional reconstruction algorithm based on super-resolution network was proposed to solve the problem that the model after CT three-dimensional reconstruction remains unclear enough in the past. The network model is a Double Loss Refinement Network （DLRNET）， and the three-dimensional reconstruction of abdominal CT was performed by uniaxial super-resolution. The optimization learning module was introduced at the end of the network model， and besides the calculation of the loss between the baseline image and super-resolution image， the loss between the roughly reconstructed image in the network model and the baseline image was also calculated. In this way， with the force of optimization learning and double loss， the results closer to the baseline image were produced by the network. Then， spatial pyramid pooling and channel attention mechanism were introduced into the feature extraction module to learn the features of vascular tissues with different thickness degrees and scales. Finally， the upsampling method was used to dynamically generate the convolution kernel set， so that a single network model was able to complete the upsampling tasks with different scaling factors. Experimental results show that compared with Residual Channel Attention Network （RCAN）， the proposed network model improves the Peak Signal-to-Noise Ratio （PSNR） by 0.789 dB on average under 2， 3， and 4 scaling factors， showing that the network model effectively improves the quality of CT three-dimensional model， recovers the continuous detail features of vascular tissues to some extent， and has practicability.

The high resolution in Magnetic Resonance （MR） image slices and low resolution between the slices lead to the lack of medical diagnostic significance of MR in the coronal and sagittal planes. In order to solve the problem， a medical image processing algorithm based on inter-layer interpolation and multi-view fusion network was proposed. Firstly， the inter-layer interpolation module was introduced to cut the MR volume data from three-dimensional data into two-dimensional images along the coronal and sagittal directions. Then， after the feature extraction on the coronal and sagittal planes， the weights were dynamically calculated by the spatial matrix filter and used for upsampling factor with any size to magnify the image. Finally， the results of the coronal and sagittal images obtained in the inter-layer interpolation module were aggregated into three-dimensional data and then cut into two-dimensional images along the axial direction. The obtained two-dimensional images were fused in pairs and corrected by the axial direction data. Experimental results show that， compared with other super-resolution algorithms， the proposed algorithm has improved the Peak Signal-to-Noise Ratio （PSNR） by about 1 dB in ×2， ×3， and ×4 scales. It can be seen that the proposed algorithm can effectively improve the quality of image reconstruction.