Normalizing CT from multiple vendors and radiation doses by routable translation network

Abstract

Quantitative evaluation of CT images from multi-site or longitudinal studies is often difficult due to the image variation depending on CT scan parameters and manufacturers. To address this problem, here we propose a novel multi-domain image translation network to convert CT images from different scan parameters and manufacturers to a normalized image. Unlike the existing multi-domain translation techniques, our method is based on the shared encoder and routable decoder architecture to maximize the expressivity and conditioning power of the network. Our network not only translates CT images between domains, but also generate a novel common space image by simply changing the routing vector. Experimental results show that the proposed CT image conversion can minimize the variation of image characteristics caused by imaging parameters, reconstruction algorithms, and hardware designs and the evaluation from the radiologists shows that the normalized images have the style of the target setting.