Super-resolution analysis of CT images based on HAT and self-integrated fusion method

In this paper, a 4-fold super-resolution analysis scheme for CT images based on HAT (Hybrid Attention Transformer) and self-integrated fusion method is proposed. The scheme realizes the effective capture and fusion of global and local information of CT images by constructing the Dense Residual Transformer Super-Resolution Analysis Architecture (DRTSR), which fuses the self-attention mechanism of Transformer and the local perception capability of Convolutional Neural Network (CNN). Meanwhile, the introduction of the self-integrated fusion method reduces the possible bias of a single model and enhances the stability and accuracy of the results by training multiple HAT models and intelligently fusing them in the testing phase. The experimental results show that the method in this paper demonstrates significant advantages in a number of evaluation indexes, and the generated super-resolution images perform better in terms of naturalness and realism, and the details and sharpness are significantly better than other methods. This work provides higher quality image data support for CT image applications and advances the field of image super-resolution analysis.