Dual-Energy CT Image Super-resolution via Generative Adversarial Network

Photon counting detector obtains CT images from multiple energy bins, and acquires X-ray intensity data of different energy bins through one X-ray exposure. However, the spatial resolution of the reconstructed image will decrease, and the image will be blurred due to the low photon count in the narrow energy box width, quantum noise and the response problem of detector cells. Deep learning is gradually applied to medical images to reduce noise or improve resolution, which has exhibited promising performance in image super-resolution (SR) by learning a nonlinear mapping function from low-resolution (LR) images to high-resolution (HR) images. Inspired by the cycle-GAN, we propose a novel network model which realize the mapping of HR images to LR images for Dual-Energy CT (DECT) reconstruction. Experimental results show that the reconstructed image has significant improvements in peak signal-to-noise ratio (PSNR) and root mean square error (RMSE). Compared with the traditional super-resolution reconstruction method, this method has better experimental results.