Self‐supervised learning of physics‐guided reconstruction neural networks without fully sampled reference data

Abstract

Purpose To develop a strategy for training a physics‐guided MRI reconstruction neural network without a database of fully sampled data sets. Methods Self‐supervised learning via data undersampling (SSDU) for physics‐guided deep learning reconstruction partitions available measurements into two disjoint sets, one of which is used in the data consistency (DC) units in the unrolled network and the other is used to define the loss for training. The proposed training without fully sampled data is compared with fully supervised training with ground‐truth data, as well as conventional compressed‐sensing and parallel imaging methods using the publicly available fastMRI knee database. The same physics‐guided neural network is used for both proposed SSDU and supervised training. The SSDU training is also applied to prospectively two‐fold accelerated high‐resolution brain data sets at different acceleration rates, and compared with parallel imaging. Results Results on five different knee sequences at an acceleration rate of 4 shows that the proposed self‐supervised approach performs closely with supervised learning, while significantly outperforming conventional compressed‐sensing and parallel imaging, as characterized by quantitative metrics and a clinical reader study. The results on prospectively subsampled brain data sets, in which supervised learning cannot be used due to lack of ground‐truth reference, show that the proposed self‐supervised approach successfully performs reconstruction at high acceleration rates (4, 6, and 8). Image readings indicate improved visual reconstruction quality with the proposed approach compared with parallel imaging at acquisition acceleration. Conclusion The proposed SSDU approach allows training of physics‐guided deep learning MRI reconstruction without fully sampled data, while achieving comparable results with supervised deep learning MRI trained on fully sampled data.

Noise2Inverse: Self-Supervised Deep Convolutional Denoising for Tomography

“…A key obstacle for the application of such techniques to large-scale 3D image reconstruction problems is their computational cost, as they involve training a new network for every 2D slice of the reconstruction. For inverse problems, approaches that rely on splitting the measurement data have recently been proposed for magnetic resonance imaging (MRI) [17], [31] and Cryo-transmission electron microscopy (Cryo-EM) [16] showing image quality improvement with respect to denoising applied on the reconstructed image. While these results are highly promising, a solid theoretical underpinning that allows analysis and insights into the interplay between the underlying noise model of the inverse problem and the obtained solution is currently lacking.…”

Improved simultaneous multislice cardiac MRI using readout concatenated k‐space SPIRiT (ROCK‐SPIRiT)

“…Higher acceleration rates in MRI have recently been achieved with physics-based deep learning methods that use algorithm unrolling to solve a regularized least-squares problem with a learned regularizer. [50][51][52][53][54][55][56] The iterative process for solving ROCK-SPIRiT naturally lends itself to such algorithm unrolling; thus, these approaches can be used to further improve regularization quality without manual parameter tuning, especially for higher acceleration rates. However, this was not explored, as we only have a limited number of such cine volumes.…”

Magnetic resonance parameter mapping using model‐guided self‐supervised deep learning

“…Recent works have also proposed the new concept of self-supervised learning for MRI reconstruction. 10,11 An early study has shown that a denoising deep learning network can be successfully trained using pairs of noisy images. 7 Self-supervised learning relies on a hypothesis that image noise and artifacts are typically incoherent in training data pairs; thus, minimizing a loss between them readily regularizes the learning to capture coherent image content.…”

“…More recently, several works have investigated unsupervised or self-supervised learning for the reconstruction of undersampled static MR images. [7][8][9][10][11][12] Although the specific implementations of these works vary from one to the other, they all train CNNs on undersampled data sets directly without fully sampled references, and inherent MR physical models (eg, Fourier encoding and coil sensitivity encoding) are incorporated as training regularizations. The results in these works have shown that with proper design of network training, unsupervised or self-supervised learning can achieve similar reconstruction performance compared with supervised learning.…”

“…The results in these works have shown that with proper design of network training, unsupervised or self-supervised learning can achieve similar reconstruction performance compared with supervised learning. 8,9,11,12 In addition to standard Fourier and coil encoding for static image reconstruction, MRI also offers a variety of physical models that are typically used in quantitative MRI and can be exploited for deep learning reconstruction of quantitative images. Examples of these models include MR relaxometry models for T 1 or T 2 mapping, diffusion models for diffusion parameter mapping, and tissue susceptibility models for quantitative susceptibility imaging.…”