ADMM-CSNet: A Deep Learning Approach for Image Compressive Sensing

Abstract

Compressive sensing (CS) is an effective approach for fast Magnetic Resonance Imaging (MRI). It aims at reconstructing MR images from a small number of under-sampled data in k-space, and accelerating the data acquisition in MRI. To improve the current MRI system in reconstruction accuracy and speed, in this paper, we propose two novel deep architectures, dubbed ADMM-Nets in basic and generalized versions. ADMM-Nets are defined over data flow graphs, which are derived from the iterative procedures in Alternating Direction Method of Multipliers (ADMM) algorithm for optimizing a general CS-based MRI model. They take the sampled k-space data as inputs and output reconstructed MR images. Moreover, we extend our network to cope with complex-valued MR images. In the training phase, all parameters of the nets, e.g., transforms, shrinkage functions, etc., are discriminatively trained end-to-end. In the testing phase, they have computational overhead similar to ADMM algorithm but use optimized parameters learned from the data for CS-based reconstruction task. We investigate different configurations in network structures and conduct extensive experiments on MR image reconstruction under different sampling rates. Due to the combination of the advantages in model-based approach and deep learning approach, the ADMM-Nets achieve state-of-the-art reconstruction accuracies with fast computational speed.

Improving resolution of MR images with an adversarial network incorporating images with different contrast

“…This aliasing artifact‐based approach showed smaller errors than the CS algorithms . Also, a learning‐based approach was applied to unfold sparse recovery framework and outperformed CS algorithms …”

“…In recent studies, deep learning algorithms, especially convolutional neural networks (CNNs), have been applied to reconstruct MR images from the incoherently down‐sampled data or to produce 7T‐like high resolution (HR) images from 3T MR images . Bahrami et al.…”

KIKI‐net: cross‐domain convolutional neural networks for reconstructing undersampled magnetic resonance images

“…We iteratively (or alternately) applied 2 different CNNs operating on different domains (the k‐space and image domain), and data consistency was interleaved among the CNNs. The second limitation of the earlier studies is that the network depth was shallower (3 to 5 layers) than that of most networks used in recent image‐restoration studies . In some studies, deep CNNs with layer depths greater than 20 afford much more promising results than shallower networks because of their larger receptive fields .…”

“…Meanwhile, several recent studies have demonstrated the applicability of deep‐learning techniques to the reconstruction of undersampled MR images or CT images . In training, tuples of undersampled images and fully sampled images are fed to convolutional neural networks (CNNs) to learn the relationship between the undersampled images and the corresponding fully sampled images .…”

“…Existing CNN‐based algorithms outperform conventional CS algorithms because of their data‐driven feature extraction and high‐nonlinearity properties . However, the existing algorithms contain 2 major limitations.…”