CLADE: Cycle Loss Augmented Degradation Enhancement for Unpaired Super-Resolution of Anisotropic Medical Images

TL;DR

CLADE addresses the lack of paired data in SRR for anisotropic 3D medical images by leveraging a CycleGAN framework with weight demodulation and a gradient-mapping cycle-consistency loss. The method learns from 2D patches in the high-resolution plane to enhance the low-resolution dimension in an unsupervised, patch-based fashion and is evaluated on abdominal MRI and CT data, where it outperforms self-supervised baselines like SMORE and reduces artefacts compared to conventional CycleGAN. Quantitative metrics (PIQUE, ES, SNR) and qualitative radiologist rankings indicate CLADE with gradient mapping achieves superior image quality and edge sharpness across orientations, while preserving SNR. The approach offers practical potential for improving visualization in clinical workflows without requiring ground-truth isotropic data, though it lacks task-specific ground-truth assessments and may benefit from broader cross-site validation and motion artifact analysis.

Abstract

Three-dimensional (3D) imaging is popular in medical applications, however, anisotropic 3D volumes with thick, low-spatial-resolution slices are often acquired to reduce scan times. Deep learning (DL) offers a solution to recover high-resolution features through super-resolution reconstruction (SRR). Unfortunately, paired training data is unavailable in many 3D medical applications and therefore we propose a novel unpaired approach; CLADE (Cycle Loss Augmented Degradation Enhancement). CLADE uses a modified CycleGAN architecture with a cycle-consistent gradient mapping loss, to learn SRR of the low-resolution dimension, from disjoint patches of the high-resolution plane within the anisotropic 3D volume data itself. We show the feasibility of CLADE in abdominal MRI and abdominal CT and demonstrate significant improvements in CLADE image quality over low-resolution volumes and state-of-the-art self-supervised SRR; SMORE (Synthetic Multi-Orientation Resolution Enhancement). Quantitative PIQUE (qualitative perception-based image quality evaluator) scores and quantitative edge sharpness (ES - calculated as the maximum gradient of pixel intensities over a border of interest), showed superior performance for CLADE in both MRI and CT. Qualitatively CLADE had the best overall image quality and highest perceptual ES over the low-resolution volumes and SMORE. This paper demonstrates the potential of using CLADE for super-resolution reconstruction of anisotropic 3D medical imaging data without the need for paired 3D training data.

Figures (16)

• Figure 1: (a) Abstracted CycleGAN model, (b) Forward cycle-consistency loss, (c) Backward cycle-consistency loss.
• Figure 2: CLADE generator architecture shown for $G_X: X \rightarrow Y$. ModConv denotes the 2D modified convolution layers. Each modified convolutional layer performs the weight-demodulation process.
• Figure 3: CLADE PatchGAN discriminator architecture.
• Figure 4: An example low-resolution sagittal MRI slice from one subject in the test data, which shows normalization errors when SRR is applied using a Conventional CycleGAN. These artifacts are removed when SRR is applied using CLADE (no $\mathcal{L}_{gmap}$). Magnified regions within the blue box are displayed beneath each image.
• Figure 5: Box Plots for MRI and CT Quantitative Metrics