Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Unrolled Meta-Learning for Multi-Coil and Multi-Modality MRI with Adaptive Optimization

Merham Fouladvand, Peuroly Batra

TL;DR

This work addresses accelerated MRI by jointly reconstructing multi-coil data and synthesizing missing modalities under undersampling. It introduces a deep unrolled framework that solves phi_eta(x) = f(x) + r_eta(x) via an adaptive forward-backward optimization with extrapolation, featuring a learnable fusion operator and complex-valued processing. The approach offers convergence guarantees and meta-learning capabilities, demonstrating superior PSNR and SSIM on fastMRI and BraTS-like tasks under aggressive undersampling and domain shifts. Overall, the method provides a scalable, generalizable solution for clinical MRI that reduces scan time while enabling robust cross-modality synthesis.

Abstract

We propose a unified deep meta-learning framework for accelerated magnetic resonance imaging (MRI) that jointly addresses multi-coil reconstruction and cross-modality synthesis. Motivated by the limitations of conventional methods in handling undersampled data and missing modalities, our approach unrolls a provably convergent optimization algorithm into a structured neural network architecture. Each phase of the network mimics a step of an adaptive forward-backward scheme with extrapolation, enabling the model to incorporate both data fidelity and nonconvex regularization in a principled manner. To enhance generalization across different acquisition settings, we integrate meta-learning, which enables the model to rapidly adapt to unseen sampling patterns and modality combinations using task-specific meta-knowledge. The proposed method is evaluated on the open source datasets, showing significant improvements in PSNR and SSIM over conventional supervised learning, especially under aggressive undersampling and domain shifts. Our results demonstrate the synergy of unrolled optimization, task-aware meta-learning, and modality fusion, offering a scalable and generalizable solution for real-world clinical MRI reconstruction.

Deep Unrolled Meta-Learning for Multi-Coil and Multi-Modality MRI with Adaptive Optimization

TL;DR

This work addresses accelerated MRI by jointly reconstructing multi-coil data and synthesizing missing modalities under undersampling. It introduces a deep unrolled framework that solves phi_eta(x) = f(x) + r_eta(x) via an adaptive forward-backward optimization with extrapolation, featuring a learnable fusion operator and complex-valued processing. The approach offers convergence guarantees and meta-learning capabilities, demonstrating superior PSNR and SSIM on fastMRI and BraTS-like tasks under aggressive undersampling and domain shifts. Overall, the method provides a scalable, generalizable solution for clinical MRI that reduces scan time while enabling robust cross-modality synthesis.

Abstract

We propose a unified deep meta-learning framework for accelerated magnetic resonance imaging (MRI) that jointly addresses multi-coil reconstruction and cross-modality synthesis. Motivated by the limitations of conventional methods in handling undersampled data and missing modalities, our approach unrolls a provably convergent optimization algorithm into a structured neural network architecture. Each phase of the network mimics a step of an adaptive forward-backward scheme with extrapolation, enabling the model to incorporate both data fidelity and nonconvex regularization in a principled manner. To enhance generalization across different acquisition settings, we integrate meta-learning, which enables the model to rapidly adapt to unseen sampling patterns and modality combinations using task-specific meta-knowledge. The proposed method is evaluated on the open source datasets, showing significant improvements in PSNR and SSIM over conventional supervised learning, especially under aggressive undersampling and domain shifts. Our results demonstrate the synergy of unrolled optimization, task-aware meta-learning, and modality fusion, offering a scalable and generalizable solution for real-world clinical MRI reconstruction.
Paper Structure (8 sections, 2 equations, 18 figures, 1 algorithm)

This paper contains 8 sections, 2 equations, 18 figures, 1 algorithm.

Table of Contents

  1. Motivation
  2. Introduction
  3. Related Work
  4. Method
  5. Experiment
  6. result
  7. Comparing masks
  8. Conclusion

Figures (18)

  • Figure 1: Uniform Cartesian Mask. Sampling Ratio 31.56% , this is the mask we used in the training. The paper we submitted also used this mask. I added more data and this is the newest result: Phase5, Avg REC PSNR is 41.7114 dB, Avg relative error is 0.026917 dB, Avg ssim is 0.9719 dB
  • Figure 2: Uniform Cartesian Mask. Sampling Ratio 62.81% , Reconstruction uses the weights of phase 5 (same as fig1), Avg REC PSNR is 26.3405 dB, Avg relative error is 0.158550 dB, Avg ssim is 0.8257
  • Figure 3: Uniform Cartesian Mask. Sampling Ratio 51.56% , Reconstruction uses the weights of phase 5 (same as fig1), Avg REC PSNR is 29.8603 dB, Avg relative error is 0.105742 dB, Avg ssim is 0.8982
  • Figure 4: Uniform Cartesian Mask. Sampling Ratio 30.63% , Reconstruction uses the weights of phase 5 (same as fig1), Avg REC PSNR is 34.2160 dB, Avg relative error is 0.063928 dB, Avg ssim is 0.9083
  • Figure 5: Uniform Cartesian Mask. Sampling Ratio 21.25% , Reconstruction uses the weights of phase 5 (same as fig1), Avg REC PSNR is 32.2787 dB, Avg relative error is 0.079973 dB, Avg ssim is 0.8814
  • ...and 13 more figures