Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Green Learning Approach to LDCT Image Restoration

Wei Wang, Yixing Wu, C. -C. Jay Kuo

TL;DR

This work proposes a green learning (GL) approach to restore medical images without loss of generality, and experiments show that the GL method offers state-of-the-art restoration performance at a smaller model size and with lower inference complexity.

Abstract

This work proposes a green learning (GL) approach to restore medical images. Without loss of generality, we use low-dose computed tomography (LDCT) images as examples. LDCT images are susceptible to noise and artifacts, where the imaging process introduces distortion. LDCT image restoration is an important preprocessing step for further medical analysis. Deep learning (DL) methods have been developed to solve this problem. We examine an alternative solution using the Green Learning (GL) methodology. The new restoration method is characterized by mathematical transparency, computational and memory efficiency, and high performance. Experiments show that our GL method offers state-of-the-art restoration performance at a smaller model size and with lower inference complexity.

A Green Learning Approach to LDCT Image Restoration

TL;DR

This work proposes a green learning (GL) approach to restore medical images without loss of generality, and experiments show that the GL method offers state-of-the-art restoration performance at a smaller model size and with lower inference complexity.

Abstract

This work proposes a green learning (GL) approach to restore medical images. Without loss of generality, we use low-dose computed tomography (LDCT) images as examples. LDCT images are susceptible to noise and artifacts, where the imaging process introduces distortion. LDCT image restoration is an important preprocessing step for further medical analysis. Deep learning (DL) methods have been developed to solve this problem. We examine an alternative solution using the Green Learning (GL) methodology. The new restoration method is characterized by mathematical transparency, computational and memory efficiency, and high performance. Experiments show that our GL method offers state-of-the-art restoration performance at a smaller model size and with lower inference complexity.
Paper Structure (9 sections, 6 figures)

This paper contains 9 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Review of Related Work
  3. Proposed GUSL Method
  4. EXPERIMENTS
  5. Experimental Setup
  6. Quality Performance Comparison
  7. Model Size and Complexity Comparison
  8. Feature Analysis
  9. Conclusion

Figures (6)

  • Figure 1: The system diagram of the proposed GUSL system. The gray boxes represent the process steps. The colored boxes stand for different types of images, with series of variant spatial sizes and shifting brightness indicating the successively changing resolutions and progressively restored coarse-to-fine information.
  • Figure 2: Visualization comparison of (a) LDCT source images, (b) NDCT source images, (c) RED-CNN restored images, (d) CTformer restored images, and (e) GUSL (ours) restored images.
  • Figure 3: A comparison of GUSL and five benchmarking methods on L506 in terms of PSNR, SSIM, the model parameter count, and the multiplier and addition counts (MACs) per pixel, where red and blue colors indicate the best and the second-best results for each metric.
  • Figure 4: The sorted RFT loss for the raw representations obtained from unsupervised representation learning at resolution Level 1.
  • Figure 5: The training-validation joint RFT ranking for the raw representations obtained from unsupervised representation learning at resolution Level 1.
  • ...and 1 more figures