Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Delta-WKV: A Novel Meta-in-Context Learner for MRI Super-Resolution

Rongchang Lu, Bingcheng Liao, Haowen Hou, Jiahang Lv, Xin Hai

TL;DR

Delta-WKV introduces a linear Transformer tailored for MRI super-resolution by fusing Meta-in-Context Learning (MiCL) with the Delta rule to dynamically adjust weights during inference, enabling efficient capture of local and global patterns without state-space modeling. The architecture blends quad-directional scanning, spatial-mixing linear attention, and channel-mixing blocks within an NRCUD-based SR backbone. It achieves state-of-the-art PSNR/SSIM on IXI and fastMRI for 2x and 4x upsampling, while delivering 15–28% reductions in training and inference time and maintaining a small parameter count (~3.0M). The work suggests strong clinical potential for large-scale, high-resolution MRI with lower computational costs and proposes future extensions to diffusion models and other imaging modalities.

Abstract

Magnetic Resonance Imaging (MRI) Super-Resolution (SR) addresses the challenges such as long scan times and expensive equipment by enhancing image resolution from low-quality inputs acquired in shorter scan times in clinical settings. However, current SR techniques still have problems such as limited ability to capture both local and global static patterns effectively and efficiently. To address these limitations, we propose Delta-WKV, a novel MRI super-resolution model that combines Meta-in-Context Learning (MiCL) with the Delta rule to better recognize both local and global patterns in MRI images. This approach allows Delta-WKV to adjust weights dynamically during inference, improving pattern recognition with fewer parameters and less computational effort, without using state-space modeling. Additionally, inspired by Receptance Weighted Key Value (RWKV), Delta-WKV uses a quad-directional scanning mechanism with time-mixing and channel-mixing structures to capture long-range dependencies while maintaining high-frequency details. Tests on the IXI and fastMRI datasets show that Delta-WKV outperforms existing methods, improving PSNR by 0.06 dB and SSIM by 0.001, while reducing training and inference times by over 15\%. These results demonstrate its efficiency and potential for clinical use with large datasets and high-resolution imaging.

Delta-WKV: A Novel Meta-in-Context Learner for MRI Super-Resolution

TL;DR

Delta-WKV introduces a linear Transformer tailored for MRI super-resolution by fusing Meta-in-Context Learning (MiCL) with the Delta rule to dynamically adjust weights during inference, enabling efficient capture of local and global patterns without state-space modeling. The architecture blends quad-directional scanning, spatial-mixing linear attention, and channel-mixing blocks within an NRCUD-based SR backbone. It achieves state-of-the-art PSNR/SSIM on IXI and fastMRI for 2x and 4x upsampling, while delivering 15–28% reductions in training and inference time and maintaining a small parameter count (~3.0M). The work suggests strong clinical potential for large-scale, high-resolution MRI with lower computational costs and proposes future extensions to diffusion models and other imaging modalities.

Abstract

Magnetic Resonance Imaging (MRI) Super-Resolution (SR) addresses the challenges such as long scan times and expensive equipment by enhancing image resolution from low-quality inputs acquired in shorter scan times in clinical settings. However, current SR techniques still have problems such as limited ability to capture both local and global static patterns effectively and efficiently. To address these limitations, we propose Delta-WKV, a novel MRI super-resolution model that combines Meta-in-Context Learning (MiCL) with the Delta rule to better recognize both local and global patterns in MRI images. This approach allows Delta-WKV to adjust weights dynamically during inference, improving pattern recognition with fewer parameters and less computational effort, without using state-space modeling. Additionally, inspired by Receptance Weighted Key Value (RWKV), Delta-WKV uses a quad-directional scanning mechanism with time-mixing and channel-mixing structures to capture long-range dependencies while maintaining high-frequency details. Tests on the IXI and fastMRI datasets show that Delta-WKV outperforms existing methods, improving PSNR by 0.06 dB and SSIM by 0.001, while reducing training and inference times by over 15\%. These results demonstrate its efficiency and potential for clinical use with large datasets and high-resolution imaging.

Paper Structure

This paper contains 12 sections, 11 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Novel Linear Transformer for MRI
  4. Model Architecture
  5. Spatial Mixing Module
  6. Channel Mixing
  7. Experiments
  8. Datasets and Implementation Details
  9. Qualitative evaluations
  10. Quantitative evaluations
  11. Ablation Study
  12. Conclusion

Figures (3)

  • Figure 1: The NRCUD block for MRI super-resolution normalizes the input image, extracts hierarchical features via multi-stage CNN filtering (Conv1 and Conv2), reconstructs high-resolution features using the SR Reconstruction module, upsamples to restore spatial dimensions, and denormalizes to generate the final high-resolution output. Our proposed Delta-WKV also adopts this structure.
  • Figure 2: The architecture of Delta-WKV.
  • Figure 3: Qualitative results with different methods under fastMRI and IXI dataset