PIE-ARNet: Prior Image Enhanced Artifact Removal Network for Limited-Angle DECT

Dual-energy computed tomography (DECT) is of great clinical significance because it can simultaneously visualize the internal structure of the scanned object and provide material-specific information. DECT obtains two attenuation measurements of the same object at two different X-ray spectra, resulting in obvious redundant information. In this context, this article suggests acquiring dual-energy projection data using two complementary incomplete scans and utilizes a pretrained Prior-Net to generate the artifact-free prior image. Then the prior image is fed into the proposed prior image enhanced artifact removal network (PIE-ARNet) together with the degraded DECT images to improve the artifact removal performance. The generator of PIE-ARNet has two encoders and two decoders, with each component being responsible for a specific task. Two encoders extract and fuse prior information and image features, while two decoders perform differential learning for data in different energy channels. The discriminator of PIE-ARNet is dedicated to transferring the real statistical properties to the generated images, producing results with enhanced visual perception. Please note that Prior-Net could be trained using the freely available conventional single-energy CT data, which will not bring extra demand for DECT data. Experiments based on the simulated data and real rat data have demonstrated the promising performance of the proposed PIE-ARNet in removing artifacts, recovering image details, and preserving reconstruction accuracy.