HepatoGEN: Generating Hepatobiliary Phase MRI with Perceptual and Adversarial Models

Jens Hoogea,*, Gerard Sanroma-Guella, Faidra Stavropouloua, Alexander Ullmanna, Gesine Knoblocha, Mark Klemensa, Carola Schmidtb Sabine Weckbacha Andreas Bolza
aBayer AG , Muellerstrasse 178, 13353 Berlin, GERMANY, bBayer Inc., 2920 Matheson Boulevard East, Mississauga, CANADA
Paper arXiv Code (Coming Soon...)

Abstract

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a crucial role in the detection and characterization of focal liver lesions, with the hepatobiliary phase (HBP) providing essential diagnostic information. However, acquiring HBP images requires prolonged scan times, which may compromise patient comfort and scanner throughput.

In this study, we propose a deep learning based approach for synthesizing HBP images from earlier contrast phases (precontrast and transitional) and compare three generative models: a perceptual U-Net, a perceptual GAN (pGAN), and a denoising diffusion probabilistic model (DDPM). We curated a multi-site DCE-MRI dataset from diverse clinical settings and introduced a contrast evolution score (CES) to assess training data quality, enhancing model performance. Quantitative evaluation using pixel-wise and perceptual metrics, combined with qualitative assessment through blinded radiologist reviews, showed that pGAN achieved the best quantitative performance but introduced heterogeneous contrast in out-of-distribution cases. In contrast, the U-Net produced consistent liver enhancement with fewer artifacts, while DDPM underperformed due to limited preservation of fine structural details.

These findings demonstrate the feasibility of synthetic HBP image generation as a means to reduce scan time without compromising diagnostic utility, highlighting the clinical potential of deep learning for dynamic contrast enhancement in liver MRI.

Dynamic Phase Liver MRI Protocol

Dynamic contrast-enhanced MRI captures the evolving uptake of contrast agent across several phases. While early phases like arterial and portal venous highlight vascular structures within the first minute, it takes up to 20 minutes before the liver parenchyma and bile ducts fully enhance—delaying diagnostic imaging.

Protocol Shortening

Our generative model leverages earlier phases (precontrast and transitional) to synthesize high-quality hepatobiliary images, aiming to reduce total scan time, without compromising diagnostic quality.

Synthesizing the Hepatobiliary Phase: Model Comparison

We generated synthetic hepatobiliary phase (HBP) images of liver MRI using three generative architectures: a perceptual U-Net, a perceptual GAN (pGAN), and a denoising diffusion probabilistic model (DDPM), applied in a slice-wise fashion. Both U-Net and pGAN produced synthetic images with contrast enhancement and anatomical detail comparable to acquired HBP scans, making them viable for diagnostic use. In contrast, DDPM consistently over-enhanced liver signal and exhibited pronounced inconsistencies across axial slices, particularly in regions with high inter-patient variability.

Synthetic Lesion Appearance in the Hepatobiliary Phase

We highlight the ability of generative models to reproduce fine anatomical details in hepatobiliary phase (HBP) MRI, which are critical for diagnosing focal liver disease. Using image comparison sliders, you can interactively compare real and synthetic HBP images for each model. We present three illustrative examples: a hepatocellular carcinoma (HCC), a focal nodular hyperplasia (FNH) lesion, and hepatic blood vessels. These cases demonstrate how well the models preserve lesion contrast, shape, and the branching structure of vessels—features essential for accurate clinical interpretation.

Focal Nodular Hyperplasia

FNH is a benign liver lesion composed of normal hepatocytes and functioning bile ducts. In HBP imaging, FNH typically appears mildly hyperintense due to uptake of hepatocyte-specific contrast agents, often featuring a central hypointense scar.

Hepatocellular Carcinoma

HCC is a malignant tumor arising from hepatocytes. In the HBP, HCC lesions are generally hypointense due to impaired uptake of contrast agents, with well-defined margins and often surrounded by a cirrhotic liver background.

Blood Vessels

Blood vessels play a crucial role in the diagnosis of focal liver disease, as many lesions exhibit characteristic vascular patterns—such as arterial hyperenhancement or portal venous washout—that aid in differentiating benign from malignant findings. Clear vessel delineation is therefore essential for reliable diagnostic interpretation.

BibTeX


      @misc{hooge2025hepatogengeneratinghepatobiliaryphase,
        title={HepatoGEN: Generating Hepatobiliary Phase MRI with Perceptual and Adversarial Models}, 
        author={Jens Hooge and Gerard Sanroma-Guell and Faidra Stavropoulou and Alexander Ullmann and Gesine Knobloch and Mark Klemens and Carola Schmidt and Sabine Weckbach and Andreas Bolz},
        year={2025},
        eprint={2504.18405},
        archivePrefix={arXiv},
        primaryClass={eess.IV},
        url={https://arxiv.org/abs/2504.18405}}