Split-slice training and hyperparameter tuning of RAKI networks for simultaneous multi-slice reconstruction

Iterative training of robust k-space interpolation networks for improved image reconstruction with limited scan specific training samples

PURPOSE

To evaluate an iterative learning approach for enhanced performance of robust artificial-neural-networks for k-space interpolation (RAKI), when only a limited amount of training data (auto-calibration signals [ACS]) are available for accelerated standard 2D imaging.

METHODS

In a first step, the RAKI model was tailored for the case of limited training data amount. In the iterative learning approach (termed iterative RAKI [iRAKI]), the tailored RAKI model is initially trained using original and augmented ACS obtained from a linear parallel imaging reconstruction. Subsequently, the RAKI convolution filters are refined iteratively using original and augmented ACS extracted from the previous RAKI reconstruction. Evaluation was carried out on 200 retrospectively undersampled in vivo datasets from the fastMRI neuro database with different contrast settings.

RESULTS

For limited training data (18 and 22 ACS lines for R = 4 and R = 5, respectively), iRAKI outperforms standard RAKI by reducing residual artifacts and yields better noise suppression when compared to standard parallel imaging, underlined by quantitative reconstruction quality metrics. Additionally, iRAKI shows better performance than both GRAPPA and standard RAKI in case of pre-scan calibration with varying contrast between training- and undersampled data.

CONCLUSION

RAKI benefits from the iterative learning approach, which preserves the noise suppression feature, but requires less original training data for the accurate reconstruction of standard 2D images thereby improving net acceleration.

Residual RAKI: A hybrid linear and non-linear approach for scan-specific k-space deep learning

Parallel imaging is the most clinically used acceleration technique for magnetic resonance imaging (MRI) in part due to its easy inclusion into routine acquisitions. In k-space based parallel imaging reconstruction, sub-sampled k-space data are interpolated using linear convolutions. At high acceleration rates these methods have inherent noise amplification and reduced image quality. On the other hand, non-linear deep learning methods provide improved image quality at high acceleration, but the availability of training databases for different scans, as well as their interpretability hinder their adaptation. In this work, we present an extension of Robust Artificial-neural-networks for k-space Interpolation (RAKI), called residual-RAKI (rRAKI), which achieves scan-specific machine learning reconstruction using a hybrid linear and non-linear methodology. In rRAKI, non-linear CNNs are trained jointly with a linear convolution implemented via a skip connection. In effect, the linear part provides a baseline reconstruction, while the non-linear CNN that runs in parallel provides further reduction of artifacts and noise arising from the linear part. The explicit split between the linear and non-linear aspects of the reconstruction also help improve interpretability compared to purely non-linear methods. Experiments were conducted on the publicly available fastMRI datasets, as well as high-resolution anatomical imaging, comparing GRAPPA and its variants, compressed sensing, RAKI, Scan Specific Artifact Reduction in K-space (SPARK) and the proposed rRAKI. Additionally, highly-accelerated simultaneous multi-slice (SMS) functional MRI reconstructions were also performed, where the proposed rRAKI was compred to Read-out SENSE-GRAPPA and RAKI. Our results show that the proposed rRAKI method substantially improves the image quality compared to conventional parallel imaging, and offers sharper images compared to SPARK and ℓ₁-SPIRiT. Furthermore, rRAKI shows improved preservation of time-varying dynamics compared to both parallel imaging and RAKI in highly-accelerated SMS fMRI.

Deep Learning and Its Application to Function Approximation for MR in Medicine: An Overview

This article presents an overview of deep learning (DL) and its applications to function approximation for MR in medicine. The aim of this article is to help readers develop various applications of DL. DL has made a large impact on the literature of many medical sciences, including MR. However, its technical details are not easily understandable for non-experts of machine learning (ML).The first part of this article presents an overview of DL and its related technologies, such as artificial intelligence (AI) and ML. AI is explained as a function that can receive many inputs and produce many outputs. ML is a process of fitting the function to training data. DL is a kind of ML, which uses a composite of many functions to approximate the function of interest. This composite function is called a deep neural network (DNN), and the functions composited into a DNN are called layers. This first part also covers the underlying technologies required for DL, such as loss functions, optimization, initialization, linear layers, non-linearities, normalization, recurrent neural networks, regularization, data augmentation, residual connections, autoencoders, generative adversarial networks, model and data sizes, and complex-valued neural networks.The second part of this article presents an overview of the applications of DL in MR and explains how functions represented as DNNs are applied to various applications, such as RF pulse, pulse sequence, reconstruction, motion correction, spectroscopy, parameter mapping, image synthesis, and segmentation.

Which multiband factor should you choose for your resting-state fMRI study?

Multiband acquisition, also called simultaneous multislice, has become a popular technique in resting-state functional connectivity studies. Multiband (MB) acceleration leads to a higher temporal resolution but also leads to spatially heterogeneous noise amplification, suggesting the costs may be greater in areas such as the subcortex. We evaluate MB factors of 2, 3, 4, 6, 8, 9, and 12 with 2 mm isotropic voxels, and additionally 2 mm and 3.3 mm single-band acquisitions, on a 32-channel head coil. Noise amplification was greater in deeper brain regions, including subcortical regions. Correlations were attenuated by noise amplification, which resulted in spatially varying biases that were more severe at higher MB factors. Temporal filtering decreased spatial biases in correlations due to noise amplification, but also tended to decrease effect sizes. In seed-based correlation maps, left-right putamen connectivity and thalamo-motor connectivity were highest in the single-band 3.3 mm protocol. In correlation matrices, MB 4, 6, and 8 had a greater number of significant correlations than the other acquisitions (both with and without temporal filtering). We recommend single-band 3.3 mm for seed-based subcortical analyses, and MB 4 provides a reasonable balance for studies analyzing both seed-based correlation maps and connectivity matrices. In multiband studies including secondary analyses of large-scale datasets, we recommend reporting effect sizes or test statistics instead of correlations. If correlations are reported, temporal filtering (or another method for thermal noise removal) should be used. The Emory Multiband Dataset is available on OpenNeuro.