Domain shift is one of the most salient challenges in medical computer\nvision. Due to immense variability in scanners' parameters and imaging\nprotocols, even images obtained from the same person and the same scanner could\ndiffer significantly. We address variability in computed tomography (CT) images\ncaused by different convolution kernels used in the reconstruction process, the\ncritical domain shift factor in CT. The choice of a convolution kernel affects\npixels' granularity, image smoothness, and noise level. We analyze a dataset of\npaired CT images, where smooth and sharp images were reconstructed from the\nsame sinograms with different kernels, thus providing identical anatomy but\ndifferent style. Though identical predictions are desired, we show that the\nconsistency, measured as the average Dice between predictions on pairs, is just\n0.54. We propose Filtered Back-Projection Augmentation (FBPAug), a simple and\nsurprisingly efficient approach to augment CT images in sinogram space\nemulating reconstruction with different kernels. We apply the proposed method\nin a zero-shot domain adaptation setup and show that the consistency boosts\nfrom 0.54 to 0.92 outperforming other augmentation approaches. Neither specific\npreparation of source domain data nor target domain data is required, so our\npublicly released FBPAug can be used as a plug-and-play module for zero-shot\ndomain adaptation in any CT-based task.\n