Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Plug-and-Play blind super-resolution of real MRI images for improved multiple sclerosis diagnosis

Matteo Cannas, Alice Mariottini, Luca Massacesi, Federica Porta, Simone Rebegoldi, Andrea Sebastiani

TL;DR

A blind super-resolution framework to enhance real 1.5 T MRI images acquired in clinical settings, where only post-processed data are available and the degradation model is not fully known is proposed.

Abstract

Magnetic resonance imaging (MRI) is central to the diagnosis of multiple sclerosis, where the identification of biomarkers such as the central vein sign benefits from high-resolution images. However, most clinical brain MRI scans are performed using 1.5 T scanners, which provide lower sensitivity compared to higher-field systems. We propose a blind super-resolution framework to enhance real 1.5 T MRI images acquired in clinical settings, where only post-processed data are available and the degradation model is not fully known. The problem is formulated as a non-convex blind inverse problem involving the joint estimation of the high-resolution image and the blur kernel. Image regularization is handled through a Plug-and-Play strategy based on a pretrained denoiser, while suitable constraints are imposed on the blur kernel. To solve the resulting model, we design a heterogeneous alternating block-coordinate method in which the two variables are updated using different types of algorithms. Convergence properties are rigorously established. Experiments on FLAIR and SWI sequences acquired at 1.5 T show improved structural definition and enhanced visibility of clinically relevant features, with visual comparison against 3 T images.

Plug-and-Play blind super-resolution of real MRI images for improved multiple sclerosis diagnosis

TL;DR

A blind super-resolution framework to enhance real 1.5 T MRI images acquired in clinical settings, where only post-processed data are available and the degradation model is not fully known is proposed.

Abstract

Magnetic resonance imaging (MRI) is central to the diagnosis of multiple sclerosis, where the identification of biomarkers such as the central vein sign benefits from high-resolution images. However, most clinical brain MRI scans are performed using 1.5 T scanners, which provide lower sensitivity compared to higher-field systems. We propose a blind super-resolution framework to enhance real 1.5 T MRI images acquired in clinical settings, where only post-processed data are available and the degradation model is not fully known. The problem is formulated as a non-convex blind inverse problem involving the joint estimation of the high-resolution image and the blur kernel. Image regularization is handled through a Plug-and-Play strategy based on a pretrained denoiser, while suitable constraints are imposed on the blur kernel. To solve the resulting model, we design a heterogeneous alternating block-coordinate method in which the two variables are updated using different types of algorithms. Convergence properties are rigorously established. Experiments on FLAIR and SWI sequences acquired at 1.5 T show improved structural definition and enhanced visibility of clinically relevant features, with visual comparison against 3 T images.
Paper Structure (10 sections, 4 theorems, 61 equations, 6 figures, 1 algorithm)

This paper contains 10 sections, 4 theorems, 61 equations, 6 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. The proposed model
  3. The proposed approach
  4. Convergence properties
  5. Numerical experiments
  6. Implementation details
  7. Model parameters.
  8. Algorithm parameters
  9. Numerical results.
  10. Conclusions

Key Result

Lemma 3.2

Let Assumption ass:1 hold. Suppose that the sequence $\{(x_k,\theta_k)\}_{k\in\mathbb{N}}$ generated by Algorithm alg:1 is bounded. Define the sequence $\{z_k\}_{k\in\mathbb{N}}\subseteq \mathbb{R}^n\times \mathbb{R}^n$ as as well as the sequence $\{\bar{z}_k\}_{k\in\mathbb{N}} \subseteq \mathbb{R}^n\times \mathbb{R}^n\times \mathbb{R}^p$ given by If $\rho<\frac{1}{2L_\phi}$ and $\alpha_x<\frac{

Figures (6)

  • Figure 1: Example of super-resolution of a FLAIR image. Two white matter hyperintense (i.e., bright) lesions are highlighted by the blue circle. The sharpness of the lesions' borders and of the cortex is remarkably improved by our approach (c) compared to the native 1.5 T FLAIR (b), offering a visual output very similar to the 3T FLAIR (a).
  • Figure 2: Example of super-resolution of a SWI image corresponding to the same brain slice reported in the FLAIR image from Figure \ref{['fig:flair_lesion1']}. A perivenular white matter lesion is highlighted by the blue circle. A small vein, appearing as a thin hypointense (i.e., black) line, is well demarcated in the 3T SWI (a), whereas it cannot be clearly identified in the 1.5T SWI (b). The super-resolution of the 1.5T SWI with our approach improves the visualization of the central vein (c).
  • Figure 3: Example of super-resolution of a FLAIR image. One white matter hyperintense (i.e., bright) lesion is highlighted by the blue circle. The sharpness of the lesion' borders and of the cortex is remarkably improved by our approach (c) compared to the native 1.5 T FLAIR (b), offering a visual output very similar to the 3T FLAIR (a).
  • Figure 4: Example of super-resolution of a SWI image corresponding to the same brain slice reported in the FLAIR image from Figure \ref{['fig:swi_lesion2']}. A perivenular white matter lesion is highlighted by the blue circle. A small vein, appearing as a thin hypointense (i.e., black) line, is well demarcated in the 3T SWI (a), whereas it cannot be clearly identified in the 1.5T SWI (b). The super-resolution of the 1.5T SWI with our approach improves the visualization of the central vein (c).
  • Figure 5: Example of super-resolution of a FLAIR image. One white matter hyperintense (i.e., bright) lesion is highlighted by the blue circle. The sharpness of the lesion' borders and of the cortex is remarkably improved by our approach (c) compared to the native 1.5 T FLAIR (b), offering a visual output very similar to the 3T FLAIR (a).
  • ...and 1 more figures

Theorems & Definitions (10)

  • Remark 3.1
  • Lemma 3.2
  • proof
  • Lemma 3.3
  • proof
  • Lemma 3.4
  • proof
  • Theorem 3.5
  • proof
  • Remark 3.6