Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Implicit Regression in Subspace for High-Sensitivity CEST Imaging

Chu Chen, Yang Liu, Se Weon Park, Jizhou Li, Kannie W. Y. Chan, Raymond H. F. Chan

TL;DR

This work addresses the low SNR limitation of Chemical Exchange Saturation Transfer (CEST) MRI by introducing Implicit Regression in Subspace (IRIS), an unsupervised denoising framework that exploits a low-dimensional subspace of z-spectra. By modeling the noise-free data as x = u v and solving a subspace regression problem, IRIS uses an implicit neural representation (RegressionNet) to learn a continuous mapping from spatial coordinates to the subspace coefficients, enabling reconstruction of denoised data x_hat = u v. Across synthetic phantoms and in-vivo mouse data, IRIS outperforms traditional and deep-learning denoising methods in MSE, PSNR, and SSIM while preserving CEST signals such as amide proton transfer (APT) and relayed NOE (rNOE), improving tumor visualization and potential clinical utility. The method yields high-sensitivity CEST maps from noisy data, which can enhance tissue characterization and assist in tumor diagnosis and treatment monitoring, with future work focused on improving computational efficiency.

Abstract

Chemical Exchange Saturation Transfer (CEST) MRI demonstrates its capability in significantly enhancing the detection of proteins and metabolites with low concentrations through exchangeable protons. The clinical application of CEST, however, is constrained by its low contrast and low signal-to-noise ratio (SNR) in the acquired data. Denoising, as one of the post-processing stages for CEST data, can effectively improve the accuracy of CEST quantification. In this work, by modeling spatial variant z-spectrums into low-dimensional subspace, we introduce Implicit Regression in Subspace (IRIS), which is an unsupervised denoising algorithm utilizing the excellent property of implicit neural representation for continuous mapping. Experiments conducted on both synthetic and in-vivo data demonstrate that our proposed method surpasses other CEST denoising methods regarding both qualitative and quantitative performance.

Implicit Regression in Subspace for High-Sensitivity CEST Imaging

TL;DR

This work addresses the low SNR limitation of Chemical Exchange Saturation Transfer (CEST) MRI by introducing Implicit Regression in Subspace (IRIS), an unsupervised denoising framework that exploits a low-dimensional subspace of z-spectra. By modeling the noise-free data as x = u v and solving a subspace regression problem, IRIS uses an implicit neural representation (RegressionNet) to learn a continuous mapping from spatial coordinates to the subspace coefficients, enabling reconstruction of denoised data x_hat = u v. Across synthetic phantoms and in-vivo mouse data, IRIS outperforms traditional and deep-learning denoising methods in MSE, PSNR, and SSIM while preserving CEST signals such as amide proton transfer (APT) and relayed NOE (rNOE), improving tumor visualization and potential clinical utility. The method yields high-sensitivity CEST maps from noisy data, which can enhance tissue characterization and assist in tumor diagnosis and treatment monitoring, with future work focused on improving computational efficiency.

Abstract

Chemical Exchange Saturation Transfer (CEST) MRI demonstrates its capability in significantly enhancing the detection of proteins and metabolites with low concentrations through exchangeable protons. The clinical application of CEST, however, is constrained by its low contrast and low signal-to-noise ratio (SNR) in the acquired data. Denoising, as one of the post-processing stages for CEST data, can effectively improve the accuracy of CEST quantification. In this work, by modeling spatial variant z-spectrums into low-dimensional subspace, we introduce Implicit Regression in Subspace (IRIS), which is an unsupervised denoising algorithm utilizing the excellent property of implicit neural representation for continuous mapping. Experiments conducted on both synthetic and in-vivo data demonstrate that our proposed method surpasses other CEST denoising methods regarding both qualitative and quantitative performance.
Paper Structure (14 sections, 9 equations, 5 figures, 2 tables)

This paper contains 14 sections, 9 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Subspace Formulation
  4. Implicit Regression
  5. Experiments
  6. Data Setup
  7. Evaluations
  8. Evaluation Metrics
  9. Synthetic Phantom
  10. In-vivo Data
  11. Ablation Studies
  12. Conclusion
  13. Acknowledgements
  14. Compliance with Ethical Standards

Figures (5)

  • Figure 1: IRIS Framework. Input CEST data is decomposed into spatial coefficients and temporal basis by SVD. RegressionNet predicts the coefficients for each position based on the coordinates input and is optimized through a loss function calculated by comparing predicted coefficients with those from SVD. Denoised CEST data is then reconstructed from predicted coefficients and temporal basis.
  • Figure 2: lnTMSE shown in boxplots indicates the median and inter-quartile range.
  • Figure 3: Denoised phantom data and signals located in different phantoms indicated by color dots.
  • Figure 4: APT (upper) and rNOE (lower) mapping from clean subject with additive noise ($\sigma=0.05$). The red arrow in T2w image indicates the tumor region.
  • Figure 5: APT (upper) and rNOE (lower) Mapping from noisy subject. The red arrow in T2w image indicates the tumor region.