Hepatobiliary phase images synthesis from multi-phase dynamic contrast-enhanced liver MR images using generative adversarial network

Motivation: Hepatobiliary Phase (HBP) of Gd-EOB-DTPA-enhanced MRI is helpful for the detection and diagnosis of liver lesions but requires waiting 20 minutes after injection of contrast agent to obtain it. Goal(s): Therefore, we aimed to use deep learning to synthesize HBP to obtain HBP images more conveniently and efficiently. Approach: We used generative adversarial network to synthesize HBP from multi-phase dynamic contrast-enhanced images. Results: The results showed that the synthetic HBP images closely mimicked the real HBP images in quantitative and qualitative image analysis, which illustrated that the model could be used to synthesize HBP in clinic to shorten the acquisition time. Impact: This study proposed a more conveniently and efficiently method to obtain hepatobiliary phase images based on generative adversarial network, which can reduce the clinical burden.