Deep Learning Based Beam Hardening Artifact Reduction in Industrial X-ray CT

In the nondestructive detection with industrial CT, due to the fact that the actual X-ray source has a wide spectrum, slices reconstructed by most existing reconstruction algorithms will suffer from beam hardening artifacts. It will degrade image quality greatly, affecting important CT image task such as CT diagnosis and so on. In this study, we propose a method to suppress beam hardening artifacts based on deep learning. We train a convolutional neural network using a large number of images with beam hardening artifacts as input and the corresponding artifact-free images reconstructed at a fixed energy as output to establish the mapping between image with beam hardening artifacts and artifact-free image for suppressing beam hardening artifacts. Experimental results show the effectiveness of the proposed method in the beam hardening artifact reduction of CT images.