Cross-platform generalization of deep learning-based denoising reconstruction algorithms for industrial computed tomography applications

The purpose of this work is to expand the application of deep learning (DL) denoising image reconstruction algorithms to industrial computed tomography (CT) scanners. Leveraging a single stage and a two-stage U-Net architectures with optional synthetic prior and transfer learning, we demonstrate robust noise reduction and artifact suppression. Validation was performed using several samples scanned with industrial CT systems, achieving significant improvements in image quality while enabling up to a 10-fold reduction in scan duration. These results underscore the versatility and adaptability of these DL reconstruction algorithms to new CT platforms, minimizing the need for scanner-specific reconstruction algorithm development and enabling its broader adoption across diverse CT imaging platforms.