Deep MR parametric imaging with the learned L + S model and attention mechanism

Abstract

Magnetic resonance (MR) parametric imaging can help the assessment of some certain diseases with its various contrast mechanisms. However, the main issue of MR parametric imaging is the long acquisition time, introducing many problems such as uncomfortable experiences for patients and motion artefacts. With the deep learning methods developing, some unrolling ones have been introduced to MR imaging as solutions. The purpose of this study is to improve both the quality and the speed of parametric imaging. The proposed method, named as LSA-Net, introduced a learned low rank plus sparsity model with attention mechanism to reconstruct highly under-sampled data for MR parametric imaging. The L+S model was used to represent the shared image structure among the parameterweighted images as low-rank part and the difference among images and the ideal model as sparse part. Besides, the attention block was introduced to effectively improved quality for the region of interest, which was most concerned in clinical applications. Then the parameter map was generated from the reconstructed parameter-weighted images by its exponential model. The LSA-Net was evaluated on the in-vivo 3D T 1𝜌 mapping, showing better performance on both image quality and time consumed than the comparing methods including LLR, SCOPE, L+S, and L+S-Net. INTRODUCTIONMR parametric imaging is a promising technique to detect subtle microscopic damage with various contrast mechanisms, enabling early diagnosis of diseases [1, 2]. Take T 1𝜌 as an exam-This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Joint Image Reconstruction and Super-Resolution for Accelerated Magnetic Resonance Imaging

“…The main drawback of the data-driven approach is its lack of interpretability, as it does not consider the problem from the perspective of the MR imaging model. Model-driven methods then gained lots of attention in MR image reconstruction [ 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 ]. Model-based reconstruction algorithms have inspired network innovations in model-driven methods to make them more flexible and interpretable [ 38 ].…”

“…ADMM-CSNet [ 18 ] unrolls the alternating direction multiplier method (ADMM) [ 39 ] for compressed sensing applications. Advanced networks, such as generative adversarial networks (GAN) [ 40 ] and attention mechanisms, are also employed in MR image reconstruction [ 14 , 24 , 25 , 26 , 27 ].…”

“…Specifically, DL has gained popularity in image restoration tasks and has shown great potential in enhancing image quality. In MR reconstruction, neural networks are employed to eliminate undersampled artifacts in the image domain or utilized as regularization in an iterative algorithm with end-to-end training [ 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 ]. For SR, neural networks learn the mapping between LR and HR image pairs [ 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ].…”