Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning-Based MR Image Re-parameterization

Abhijeet Narang, Abhigyan Raj, Mihaela Pop, Mehran Ebrahimi

TL;DR

This work tackles MRI re-parameterization, aiming to synthesize MR scans under new acquisition parameters to avoid multiple scans. It introduces a two-stage deep learning approach: an image reconstruction auto-encoder for extracting robust features and a coarse-to-fine Param-Net that maps input parameters to a new image at the desired settings, with two training regimes—Default-to-Param and Param-to-Param. Evaluations on synthetic data from MRiLab and BrainWeb show the Default-to-Param model generally outperforming Param-to-Param in PSNR and MAE, indicating the approach can learn non-linear re-parameterizations more efficiently than traditional biophysical models. The method offers faster generation of parameter-conditioned MR images and has potential for extending to broader parameter spaces, including $T1$/$T2$ mapping, given sufficient training data.

Abstract

Magnetic resonance (MR) image re-parameterization refers to the process of generating via simulations of an MR image with a new set of MRI scanning parameters. Different parameter values generate distinct contrast between different tissues, helping identify pathologic tissue. Typically, more than one scan is required for diagnosis; however, acquiring repeated scans can be costly, time-consuming, and difficult for patients. Thus, using MR image re-parameterization to predict and estimate the contrast in these imaging scans can be an effective alternative. In this work, we propose a novel deep learning (DL) based convolutional model for MRI re-parameterization. Based on our preliminary results, DL-based techniques hold the potential to learn the non-linearities that govern the re-parameterization.

Deep Learning-Based MR Image Re-parameterization

TL;DR

This work tackles MRI re-parameterization, aiming to synthesize MR scans under new acquisition parameters to avoid multiple scans. It introduces a two-stage deep learning approach: an image reconstruction auto-encoder for extracting robust features and a coarse-to-fine Param-Net that maps input parameters to a new image at the desired settings, with two training regimes—Default-to-Param and Param-to-Param. Evaluations on synthetic data from MRiLab and BrainWeb show the Default-to-Param model generally outperforming Param-to-Param in PSNR and MAE, indicating the approach can learn non-linear re-parameterizations more efficiently than traditional biophysical models. The method offers faster generation of parameter-conditioned MR images and has potential for extending to broader parameter spaces, including / mapping, given sufficient training data.

Abstract

Magnetic resonance (MR) image re-parameterization refers to the process of generating via simulations of an MR image with a new set of MRI scanning parameters. Different parameter values generate distinct contrast between different tissues, helping identify pathologic tissue. Typically, more than one scan is required for diagnosis; however, acquiring repeated scans can be costly, time-consuming, and difficult for patients. Thus, using MR image re-parameterization to predict and estimate the contrast in these imaging scans can be an effective alternative. In this work, we propose a novel deep learning (DL) based convolutional model for MRI re-parameterization. Based on our preliminary results, DL-based techniques hold the potential to learn the non-linearities that govern the re-parameterization.
Paper Structure (13 sections, 6 figures, 2 tables)

This paper contains 13 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. AutoEncoder: Feature Extraction
  5. Param-Net
  6. Default-to-Param
  7. Param-to-Param
  8. Dataset
  9. Network Training
  10. Results
  11. MRiLab Testset
  12. Brainweb Testset
  13. Conclusion

Figures (6)

  • Figure 1: Autoencoder
  • Figure 2: Param-Net
  • Figure 3: Re-parameterization results for Default-to-Param Model, signal intensity (SI) and the absolute difference are represented in arbitrary units.
  • Figure 4: Re-parameterization results for Param-to-Param Model, signal intensity (SI) and the absolute difference are represented in arbitrary units.
  • Figure 5: Re-parameterization results for Default-to-Param Model, signal intensity (SI) and the absolute difference are represented in arbitrary units.
  • ...and 1 more figures