Monoplanar CT Reconstruction with GANs

Reconstructing Computed Tomography images (CT) from radiographs currently requires biplanar radiographs for accurate CT reconstruction due to the complementary information contained in the individual views. However, in many cases biplanar information is not available. In this work, we therefore propose a KNN and a PCA-based approach using biplanar radiographs only at the training stage while performing the final inference using only a single anterior-posterior radiograph, thereby increasing the applicability and usability of the model. The methods are quantitatively validated on a multiview database achieving 81% PSNR of biplanar inference and also qualitatively on a dataset of radiographs with no corresponding CT scans.