CT Images Super-Resolution Reconstruction Using Bi-level Routing Attention and Consecutive Dilated Convolutions

The use of low-dose CT results in lower radiation exposure and mitigates the effects on the human body. Low-dose computed tomography imaging can be hampered by noise and artefacts, resulting in inconclusive diagnostic results. These problems can hinder effective patient treatment and management. In order to address this matter, the present study endeavors to introduce an approach aiming at super-resolution reconstruction of low-dose CT images. This technique employs a Bi-Level Routing Attention (BRA) module and a Consecutive Dilated Convolutions (CDC) module to enhance the realism of CT images The method incorporates the BRA and CDC modules with the U-Net neural network. The CDC module has the ability to acquire more complex features that exist on multiple scales. Similarly, the BRA module incorporates detailed background information into these features, which enhances the reconstruction process's overall resolution. This method is intended to enhance the module's ability to retrieve image features. Compared with mainstream super-resolution reconstruction algorithms, this model improves both objective and subjective evaluation metrics. It can recover image detail more clearly and reconstruct better quality.