An Optimization-Based Meta-Learning Model for MRI Reconstruction with Diverse Dataset

Abstract

(1) Purpose: This work aims at developing a generalizable MRI reconstruction model in the meta-learning framework.
 The standard benchmarks in meta-learning are challenged by learning on diverse task distributions. The proposed network learns the regularization function in a variational model and reconstructs MR images with various under-sampling ratios or patterns that may or may not be seen in the training data by leveraging a heterogeneous dataset.
 (2) Methods: We propose an
 unrolling network induced by learnable optimization algorithms (LOA) for solving our nonconvex nonsmooth variational model for MRI reconstruction. In this model, the learnable regularization function contains a task-invariant common feature encoder and task-specific learner represented by a shallow network. To train the network we split the training data into two parts: training and validation, and introduce a bilevel optimization algorithm. The lower-level optimization trains task-invariant parameters for the feature encoder with fixed parameters of the task-specific learner on the training dataset, and the upper-level optimizes the parameters of the task-specific learner on the validation dataset.
 (3) Results: The PSNR increases 1.5 dB on average compared to the network trained through conventional supervised learning on the seen CS ratios. We test the result of quick adaption on the unseen tasks after meta-training, the average PSNR arises 1.22 dB compared to the conventional learning procedure that is directly trained on the unseen CS ratios in the meanwhile saving half of the training time. The average PSNR arises 1.87 dB for unseen sampling patterns comparing to conventional learning; 
 (4) Conclusion: We proposed a meta-learning framework consisting of the base network architecture, design of regularization, and bi-level optimization-based training. The network inherits the convergence property of the LOA and interpretation of the variational model. The generalization ability is improved by the designated regularization and bilevel optimization-based training algorithm.

A Learnable Variational Model for Joint Multimodal MRI Reconstruction and Synthesis

“…Denote X = {x 1 , x 2 , x 3 }. The Clark subdifferential of each x i is identical to [2] expect for an additional smooth term in (1), so the analysis for each individual x i is the same as [2]. It has been proved in [2]…”

“…As both h wi and g θ are compositions of Lipschitz continuous, we know the first and last terms of Ψ ε Θ,γ (X) are Lipschitz continuous. The second sum is Lipschitz continuous proved by the Lemma A2 in [2].…”

“…The third term is to promote high structural similarity index [21] of the reconstructed/synthesized images. In (2), the lower-level optimization learns the network parameters Θ with any fixed coefficient γ on the training set, and the upper-level tunes the hyper-parameter γ on the validation set, which mitigates the challenging overfitting issue.…”

“…We need the following lemmas to prove Theorem 1. Since Lemma 2 can be verified by direct calculation, and Lemma 3 and 4 are similar as those given in [4,2], we omit their proofs here.…”

“…To overcome the weakness of the end-toend black-box networks, several learnable optimization algorithms (LOAs) have been developed attracted much attention, which possess of a more interpretable network architecture. LOA-based reconstruction methods unfold the iterative optimization algorithm into a multi-phase network in which the regularization and image transformation are learned to improve the network performance [15,2,4,1], e.g. ADMM-Net [20], ISTA-Net + [26] and PD-Net [5].…”