Image synthesis with deep convolutional generative adversarial networks for material decomposition in dual-energy CT from a kilovoltage CT

Generative Adversarial Networks (GANs) have been widely used and it is expected to use for the clinical examination image. The objective of current study was synthesize material decomposition images bone-water (bone(water)) fat-water (fat(water)) reconstructed from dual-energy computed tomography (DECT) using an equivalent kilovoltage-CT (kV-CT) image a deep conditional GAN. effective atomic number were DECT. We 18,084 28 patients divided into two datasets: training data model included 16,146 (20 patients) test evaluation 1938 (8 patients). Image prediction frameworks single energy CT at 120 kVp created. image-synthesis framework based on CNN with generator discriminator. mean absolute error (MAE), relative square (MSE), root (RMSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), mutual information (MI) evaluated. Hounsfield unit (HU) difference between synthesized reference bone(water) fat(water) within 5.3 HU 20.3 HU, respectively. average MAE, MSE, RMSE, SSIM, MI 0.8, 1.3, 0.9, 55.3, 0.0, 0.1, 72.1, 1.4, proposed can act as suitable alternative existing methods reconstruction via DECT kV-CT.