3D Volumetric Super-Resolution in Radiology Using 3D RRDB-GAN

TL;DR

The work introduces 3D RRDB-GAN, a volumetric super-resolution framework for radiology that integrates a 2.5D perceptual loss to boost 3D image realism. By extending RRDBNet to 3D with a 3D UNet discriminator and employing k-space degradation for realistic low-resolution pairs, the method achieves 4x upscaling across diverse MRI/CT datasets. Across Mice Brain MRH, OASIS, HCP1200, and MSD-Task-6, it delivers superior perceptual metrics LPIPS and FID, though traditional metrics SSIM and PSNR show mixed results. The approach increases depth, clarity, and volumetric detail in medical images, offering potential to enhance interpretation and diagnostic workflows in 3D radiology contexts.

Abstract

This study introduces the 3D Residual-in-Residual Dense Block GAN (3D RRDB-GAN) for 3D super-resolution for radiology imagery. A key aspect of 3D RRDB-GAN is the integration of a 2.5D perceptual loss function, which contributes to improved volumetric image quality and realism. The effectiveness of our model was evaluated through 4x super-resolution experiments across diverse datasets, including Mice Brain MRH, OASIS, HCP1200, and MSD-Task-6. These evaluations, encompassing both quantitative metrics like LPIPS and FID and qualitative assessments through sample visualizations, demonstrate the models effectiveness in detailed image analysis. The 3D RRDB-GAN offers a significant contribution to medical imaging, particularly by enriching the depth, clarity, and volumetric detail of medical images. Its application shows promise in enhancing the interpretation and analysis of complex medical imagery from a comprehensive 3D perspective.