Accelerated coronary MRI with sRAKI: A database-free self-consistent neural network k-space reconstruction for arbitrary undersampling

Accelerated 3D whole-heart non-contrast-enhanced mDIXON coronary MR angiography using deep learning-constrained compressed sensing reconstruction

OBJECTIVES

To investigate the feasibility of a deep learning-constrained compressed sensing (DL-CS) method in non-contrast-enhanced modified DIXON (mDIXON) coronary magnetic resonance angiography (MRA) and compare its diagnostic accuracy using coronary CT angiography (CCTA) as a reference standard.

METHODS

Ninety-nine participants were prospectively recruited for this study. Thirty healthy subjects (age range: 20-65 years; 50% female) underwent three non-contrast mDIXON-based coronary MRA sequences including DL-CS, CS, and conventional sequences. The three groups were compared based on the scan time, subjective image quality score, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). The remaining 69 patients suspected of coronary artery disease (CAD) (age range: 39-83 years; 51% female) underwent the DL-CS coronary MRA and its diagnostic performance was compared with that of CCTA.

RESULTS

The scan time for the DL-CS and CS sequences was notably shorter than that of the conventional sequence (9.6 ± 3.1 min vs 10.0 ± 3.4 min vs 13.0 ± 4.9 min; p < 0.001). The DL-CS sequence obtained the highest image quality score, mean SNR, and CNR compared to CS and conventional methods (all p < 0.001). Compared to CCTA, the accuracy, sensitivity, and specificity of DL-CS mDIXON coronary MRA per patient were 84.1%, 92.0%, and 79.5%; those per vessel were 90.3%, 82.6%, and 92.5%; and those per segment were 98.0%, 85.1%, and 98.0%, respectively.

CONCLUSION

The DL-CS mDIXON coronary MRA provided superior image quality and short scan time for visualizing coronary arteries in healthy individuals and demonstrated high diagnostic value compared to CCTA in CAD patients.

CRITICAL RELEVANCE STATEMENT

DL-CS resulted in improved image quality with an acceptable scan time, and demonstrated excellent diagnostic performance compared to CCTA, which could be an alternative to enhance the workflow of coronary MRA.

KEY POINTS

Current coronary MRA techniques are limited by scan time and the need for noise reduction. DL-CS reduced the scan time in coronary MR angiography. Deep learning achieved the highest image quality among the three methods. Deep learning-based coronary MR angiography demonstrated high performance compared to CT angiography.

Knowledge-driven deep learning for fast MR imaging: Undersampled MR image reconstruction from supervised to un-supervised learning

Deep learning (DL) has emerged as a leading approach in accelerating MRI. It employs deep neural networks to extract knowledge from available datasets and then applies the trained networks to reconstruct accurate images from limited measurements. Unlike natural image restoration problems, MRI involves physics-based imaging processes, unique data properties, and diverse imaging tasks. This domain knowledge needs to be integrated with data-driven approaches. Our review will introduce the significant challenges faced by such knowledge-driven DL approaches in the context of fast MRI along with several notable solutions, which include learning neural networks and addressing different imaging application scenarios. The traits and trends of these techniques have also been given which have shifted from supervised learning to semi-supervised learning, and finally, to unsupervised learning methods. In addition, MR vendors' choices of DL reconstruction have been provided along with some discussions on open questions and future directions, which are critical for the reliable imaging systems.

Improving the efficiency and accuracy of cardiovascular magnetic resonance with artificial intelligence-review of evidence and proposition of a roadmap to clinical translation

BACKGROUND

Cardiovascular magnetic resonance (CMR) is an important imaging modality for the assessment of heart disease; however, limitations of CMR include long exam times and high complexity compared to other cardiac imaging modalities. Recently advancements in artificial intelligence (AI) technology have shown great potential to address many CMR limitations. While the developments are remarkable, translation of AI-based methods into real-world CMR clinical practice remains at a nascent stage and much work lies ahead to realize the full potential of AI for CMR.

METHODS

Herein we review recent cutting-edge and representative examples demonstrating how AI can advance CMR in areas such as exam planning, accelerated image reconstruction, post-processing, quality control, classification and diagnosis.

RESULTS

These advances can be applied to speed up and simplify essentially every application including cine, strain, late gadolinium enhancement, parametric mapping, 3D whole heart, flow, perfusion and others. AI is a unique technology based on training models using data. Beyond reviewing the literature, this paper discusses important AI-specific issues in the context of CMR, including (1) properties and characteristics of datasets for training and validation, (2) previously published guidelines for reporting CMR AI research, (3) considerations around clinical deployment, (4) responsibilities of clinicians and the need for multi-disciplinary teams in the development and deployment of AI in CMR, (5) industry considerations, and (6) regulatory perspectives.

CONCLUSIONS

Understanding and consideration of all these factors will contribute to the effective and ethical deployment of AI to improve clinical CMR.

Scan-Specific Self-Supervised Bayesian Deep Non-Linear Inversion for Undersampled MRI Reconstruction

Magnetic resonance imaging is subject to slow acquisition times due to the inherent limitations in data sampling. Recently, supervised deep learning has emerged as a promising technique for reconstructing sub-sampled MRI. However, supervised deep learning requires a large dataset of fully-sampled data. Although unsupervised or self-supervised deep learning methods have emerged to address the limitations of supervised deep learning approaches, they still require a database of images. In contrast, scan-specific deep learning methods learn and reconstruct using only the sub-sampled data from a single scan. Here, we introduce Scan-Specific Self-Supervised Bayesian Deep Non-Linear Inversion (DNLINV) that does not require an auto calibration scan region. DNLINV utilizes a Deep Image Prior-type generative modeling approach and relies on approximate Bayesian inference to regularize the deep convolutional neural network. We demonstrate our approach on several anatomies, contrasts, and sampling patterns and show improved performance over existing approaches in scan-specific calibrationless parallel imaging and compressed sensing.

A motion-corrected deep-learning reconstruction framework for accelerating whole-heart magnetic resonance imaging in patients with congenital heart disease

BACKGROUND

Cardiovascular magnetic resonance (CMR) is an important imaging modality for the assessment and management of adult patients with congenital heart disease (CHD). However, conventional techniques for three-dimensional (3D) whole-heart acquisition involve long and unpredictable scan times and methods that accelerate scans via k-space undersampling often rely on long iterative reconstructions. Deep-learning-based reconstruction methods have recently attracted much interest due to their capacity to provide fast reconstructions while often outperforming existing state-of-the-art methods. In this study, we sought to adapt and validate a non-rigid motion-corrected model-based deep learning (MoCo-MoDL) reconstruction framework for 3D whole-heart MRI in a CHD patient cohort.

METHODS

The previously proposed deep-learning reconstruction framework MoCo-MoDL, which incorporates a non-rigid motion-estimation network and a denoising regularization network within an unrolled iterative reconstruction, was trained in an end-to-end manner using 39 CHD patient datasets. Once trained, the framework was evaluated in eight CHD patient datasets acquired with seven-fold prospective undersampling. Reconstruction quality was compared with the state-of-the-art non-rigid motion-corrected patch-based low-rank reconstruction method (NR-PROST) and against reference images (acquired with three-or-four-fold undersampling and reconstructed with NR-PROST).

RESULTS

Seven-fold undersampled scan times were 2.1 ± 0.3 minutes and reconstruction times were ∼30 seconds, approximately 240 times faster than an NR-PROST reconstruction. Image quality comparable to the reference images was achieved using the proposed MoCo-MoDL framework, with no statistically significant differences found in any of the assessed quantitative or qualitative image quality measures. Additionally, expert image quality scores indicated the MoCo-MoDL reconstructions were consistently of a higher quality than the NR-PROST reconstructions of the same data, with the differences in 12 of the 22 scores measured for individual vascular structures found to be statistically significant.

CONCLUSION

The MoCo-MoDL framework was applied to an adult CHD patient cohort, achieving good quality 3D whole-heart images from ∼2-minute scans with reconstruction times of ∼30 seconds.

Parallel-stream fusion of scan-specific and scan-general priors for learning deep MRI reconstruction in low-data regimes

Magnetic resonance imaging (MRI) is an essential diagnostic tool that suffers from prolonged scan times. Reconstruction methods can alleviate this limitation by recovering clinically usable images from accelerated acquisitions. In particular, learning-based methods promise performance leaps by employing deep neural networks as data-driven priors. A powerful approach uses scan-specific (SS) priors that leverage information regarding the underlying physical signal model for reconstruction. SS priors are learned on each individual test scan without the need for a training dataset, albeit they suffer from computationally burdening inference with nonlinear networks. An alternative approach uses scan-general (SG) priors that instead leverage information regarding the latent features of MRI images for reconstruction. SG priors are frozen at test time for efficiency, albeit they require learning from a large training dataset. Here, we introduce a novel parallel-stream fusion model (PSFNet) that synergistically fuses SS and SG priors for performant MRI reconstruction in low-data regimes, while maintaining competitive inference times to SG methods. PSFNet implements its SG prior based on a nonlinear network, yet it forms its SS prior based on a linear network to maintain efficiency. A pervasive framework for combining multiple priors in MRI reconstruction is algorithmic unrolling that uses serially alternated projections, causing error propagation under low-data regimes. To alleviate error propagation, PSFNet combines its SS and SG priors via a novel parallel-stream architecture with learnable fusion parameters. Demonstrations are performed on multi-coil brain MRI for varying amounts of training data. PSFNet outperforms SG methods in low-data regimes, and surpasses SS methods with few tens of training samples. On average across tasks, PSFNet achieves 3.1 dB higher PSNR, 2.8% higher SSIM, and 0.3 × lower RMSE than baselines. Furthermore, in both supervised and unsupervised setups, PSFNet requires an order of magnitude lower samples compared to SG methods, and enables an order of magnitude faster inference compared to SS methods. Thus, the proposed model improves deep MRI reconstruction with elevated learning and computational efficiency.

Feasibility of accelerated non-contrast-enhanced whole-heart bSSFP coronary MR angiography by deep learning-constrained compressed sensing

OBJECTIVES

To examine a compressed sensing artificial intelligence (CSAI) framework to accelerate image acquisition in non-contrast-enhanced whole-heart bSSFP coronary magnetic resonance (MR) angiography.

METHODS

Thirty healthy volunteers and 20 patients with suspected coronary artery disease (CAD) scheduled for coronary computed tomography angiography (CCTA) were enrolled. Non-contrast-enhanced coronary MR angiography was performed with CSAI, compressed sensing (CS), and sensitivity encoding (SENSE) methods in healthy participants and with CSAI in patients. Acquisition time, subjective image quality score, and objective image quality measurement (blood pool homogeneity, signal-to-noise ratio [SNR], and contrast-to-noise ratio [CNR]) were compared among the three protocols. The diagnostic performance of CASI coronary MR angiography for predicting significant stenosis (≥ 50% diameter stenosis) on CCTA was evaluated. The Friedman test was performed to compare the three protocols.

RESULTS

Acquisition time was significantly shorter in the CSAI and CS groups than in the SENSE group (10.2 ± 3.2 min vs. 10.9 ± 2.9 min vs. 13.0 ± 4.1 min, p < 0.001). However, the CSAI approach had the highest image quality scores, blood pool homogeneity, mean SNR value, and mean CNR value (all p < 0.001) compared with the CS and SENSE approaches. The sensitivity, specificity, and accuracy of CSAI coronary MR angiography per patient were 87.5% (7/8), 91.7% (11/12), and 90.0% (18/20); those per vessel were 81.8% (9/11), 93.9% (46/49), and 91.7% (55/60); and those per segment were 84.6% (11/13), 98.0% (244/249), and 97.3% (255/262), respectively.

CONCLUSIONS

CSAI yielded superior image quality within a clinically feasible acquisition time in healthy participants and patients with suspected CAD.

CLINICAL RELEVANCE STATEMENT

The non-invasive and radiation-free CSAI framework could be a promising tool for rapid screening and comprehensive examination of the coronary vasculature in patients with suspected CAD.

KEY POINTS

• This prospective study showed that CSAI enables a reduction in acquisition time by 22% with superior diagnostic image quality compared with the SENSE protocol. • CSAI replaces the wavelet transform with a CNN as a sparsifying transform in the CS algorithm, achieving high coronary MR image quality with reduced noise. • CSAI achieved per-patient sensitivity of 87.5% (7/8) and specificity of 91.7% (11/12) respectively for detecting significant coronary stenosis.

Deep Learning-Based Acceleration of Compressed Sensing for Noncontrast-Enhanced Coronary Magnetic Resonance Angiography in Patients With Suspected Coronary Artery Disease

BACKGROUND

The clinical application of coronary MR angiography (MRA) remains limited due to its long acquisition time and often unsatisfactory image quality. A compressed sensing artificial intelligence (CSAI) framework was recently introduced to overcome these limitations, but its feasibility in coronary MRA is unknown.

PURPOSE

To evaluate the diagnostic performance of noncontrast-enhanced coronary MRA with CSAI in patients with suspected coronary artery disease (CAD).

STUDY TYPE

Prospective observational study.

POPULATION

A total of 64 consecutive patients (mean age ± standard deviation [SD]: 59 ± 10 years, 48.4% females) with suspected CAD.

FIELD STRENGTH/SEQUENCE

A 3.0-T, balanced steady-state free precession sequence.

ASSESSMENT

Three observers evaluated the image quality for 15 coronary segments of the right and left coronary arteries using a 5-point scoring system (1 = not visible; 5 = excellent). Image scores ≥3 were considered diagnostic. Furthermore, the detection of CAD with ≥50% stenosis was evaluated in comparison to reference standard coronary computed tomography angiography (CTA). Mean acquisition times for CSAI-based coronary MRA were measured.

STATISTICAL TESTS

For each patient, vessel and segment, sensitivity, specificity, and diagnostic accuracy of CSAI-based coronary MRA for detecting CAD with ≥50% stenosis according to coronary CTA were calculated. Intraclass correlation coefficients (ICCs) were used to assess the interobserver agreement.

RESULTS

The mean MR acquisition time ± SD was 8.1 ± 2.4 minutes. Twenty-five (39.1%) patients had CAD with ≥50% stenosis on coronary CTA and 29 (45.3%) patients on MRA. A total of 885 segments on the CTA images and 818/885 (92.4%) coronary MRA segments were diagnostic (image score ≥3). The sensitivity, specificity, and diagnostic accuracy were as follows: per patient (92.0%, 84.6%, and 87.5%), per vessel (82.9%, 93.4%, and 91.1%), and per segment (77.6%, 98.2%, and 96.6%), respectively. The ICCs for image quality and stenosis assessment were 0.76-0.99 and 0.66-1.00, respectively.

DATA CONCLUSION

The image quality and diagnostic performance of coronary MRA with CSAI may show good results in comparison to coronary CTA in patients with suspected CAD.

EVIDENCE LEVEL

1.

TECHNICAL EFFICACY

2.

Linear fine-tuning: a linear transformation based transfer strategy for deep MRI reconstruction

Introduction

Fine-tuning (FT) is a generally adopted transfer learning method for deep learning-based magnetic resonance imaging (MRI) reconstruction. In this approach, the reconstruction model is initialized with pre-trained weights derived from a source domain with ample data and subsequently updated with limited data from the target domain. However, the direct full-weight update strategy can pose the risk of "catastrophic forgetting" and overfitting, hindering its effectiveness. The goal of this study is to develop a zero-weight update transfer strategy to preserve pre-trained generic knowledge and reduce overfitting.

Methods

Based on the commonality between the source and target domains, we assume a linear transformation relationship of the optimal model weights from the source domain to the target domain. Accordingly, we propose a novel transfer strategy, linear fine-tuning (LFT), which introduces scaling and shifting (SS) factors into the pre-trained model. In contrast to FT, LFT only updates SS factors in the transfer phase, while the pre-trained weights remain fixed.

Results

To evaluate the proposed LFT, we designed three different transfer scenarios and conducted a comparative analysis of FT, LFT, and other methods at various sampling rates and data volumes. In the transfer scenario between different contrasts, LFT outperforms typical transfer strategies at various sampling rates and considerably reduces artifacts on reconstructed images. In transfer scenarios between different slice directions or anatomical structures, LFT surpasses the FT method, particularly when the target domain contains a decreasing number of training images, with a maximum improvement of up to 2.06 dB (5.89%) in peak signal-to-noise ratio.

Discussion

The LFT strategy shows great potential to address the issues of "catastrophic forgetting" and overfitting in transfer scenarios for MRI reconstruction, while reducing the reliance on the amount of data in the target domain. Linear fine-tuning is expected to shorten the development cycle of reconstruction models for adapting complicated clinical scenarios, thereby enhancing the clinical applicability of deep MRI reconstruction.

JSENSE-Pro: Joint sensitivity estimation and image reconstruction in parallel imaging using pre-learned subspaces of coil sensitivity functions

PURPOSE

To improve calibrationless parallel imaging using pre-learned subspaces of coil sensitivity functions.

THEORY AND METHODS

A subspace-based joint sensitivity estimation and image reconstruction method was developed for improved parallel imaging with no calibration data. Specifically, we proposed to use a probabilistic subspace model to capture prior information of the coil sensitivity functions from previous scans acquired using the same receiver system. Both the subspace basis and coefficient distributions were learned from a small set of training data. The learned subspace model was then incorporated into the regularized reconstruction formalism that includes a sparsity prior. The nonlinear optimization problem was solved using alternating minimization algorithm. Public fastMRI brain dataset was retrospectively undersampled by different schemes for performance evaluation of the proposed method.

RESULTS

With no calibration data, the proposed method consistently produced the most accurate coil sensitivity estimation and highest quality image reconstructions at all acceleration factors tested in comparison with state-of-the-art methods including JSENSE, DeepSENSE, P-LORAKS, and Sparse BLIP. Our results are comparable to or even better than those from SparseSENSE, which used calibration data for sensitivity estimation. The work also demonstrated that the probabilistic subspace model learned from T₂ w data can be generalized to aiding the reconstruction of FLAIR data acquired from the same receiver system.

CONCLUSION

A subspace-based method named JSENSE-Pro has been proposed for accelerated parallel imaging without the acquisition of companion calibration data. The method is expected to further enhance the practical utility of parallel imaging, especially in applications where calibration data acquisition is not desirable or limited.

Multi-weight respecification of scan-specific learning for parallel imaging

Parallel imaging is widely used in magnetic resonance imaging as an acceleration technology. Traditional linear reconstruction methods in parallel imaging often suffer from noise amplification. Recently, a non-linear robust artificial-neural-network for k-space interpolation (RAKI) exhibits superior noise resilience over other linear methods. However, RAKI performs poorly at high acceleration rates and needs a large number of autocalibration signals as the training samples. In order to tackle these issues, we propose a multi-weight method that implements multiple weighting matrices on the under-sampled data, named MW-RAKI. Enforcing multiple weighted matrices on the measurements can effectively reduce the influence of noise and increase the data constraints. Furthermore, we incorporate the strategy of multiple weighting matrixes into a residual version of RAKI, and form MW-rRAKI. Experimental comparisons with the alternative methods demonstrated noticeably better reconstruction performances, particularly at high acceleration rates. With only 12.5% of the k-space data is available, the PSNR of MW-RAKI and MW-rRAKI is improved by about 3 dB and 4 dB compared to RAKI and rRAKI, respectively.

Reduction in Acquisition Time and Improvement in Image Quality in T2-Weighted MR Imaging of Musculoskeletal Tumors of the Extremities Using a Novel Deep Learning-Based Reconstruction Technique in a Turbo Spin Echo (TSE) Sequence

Background: The aim of this study was to assess the technical feasibility and the impact on image quality and acquisition time of a deep learning-accelerated fat-saturated T2-weighted turbo spin echo sequence in musculoskeletal imaging of the extremities. Methods: Twenty-three patients who underwent MRI of the extremities were prospectively included. Standard T2w turbo inversion recovery magnitude (TIRMStd) imaging was compared to a deep learning-accelerated T2w TSE (TSEDL) sequence. Image analysis of 23 patients with a mean age of 60 years (range 30−86) was performed regarding image quality, noise, sharpness, contrast, artifacts, lesion detectability and diagnostic confidence. Pathological findings were documented measuring the maximum diameter. Results: The analysis showed a significant improvement for the T2 TSEDL with regard to image quality, noise, contrast, sharpness, lesion detectability, and diagnostic confidence, as compared to T2 TIRMStd (each p < 0.001). There were no differences in the number of detected lesions. The time of acquisition (TA) could be reduced by 52−59%. Interrater agreement was almost perfect (κ = 0.886). Conclusion: Accelerated T2 TSEDL was technically feasible and superior to conventionally applied T2 TIRMStd. Concurrently, TA could be reduced by 52−59%. Therefore, deep learning-accelerated MR imaging is a promising and applicable method in musculoskeletal imaging.

Unsupervised MRI Reconstruction via Zero-Shot Learned Adversarial Transformers

Supervised reconstruction models are characteristically trained on matched pairs of undersampled and fully-sampled data to capture an MRI prior, along with supervision regarding the imaging operator to enforce data consistency. To reduce supervision requirements, the recent deep image prior framework instead conjoins untrained MRI priors with the imaging operator during inference. Yet, canonical convolutional architectures are suboptimal in capturing long-range relationships, and priors based on randomly initialized networks may yield suboptimal performance. To address these limitations, here we introduce a novel unsupervised MRI reconstruction method based on zero-Shot Learned Adversarial TransformERs (SLATER). SLATER embodies a deep adversarial network with cross-attention transformers to map noise and latent variables onto coil-combined MR images. During pre-training, this unconditional network learns a high-quality MRI prior in an unsupervised generative modeling task. During inference, a zero-shot reconstruction is then performed by incorporating the imaging operator and optimizing the prior to maximize consistency to undersampled data. Comprehensive experiments on brain MRI datasets clearly demonstrate the superior performance of SLATER against state-of-the-art unsupervised methods.

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.

A review and experimental evaluation of deep learning methods for MRI reconstruction

Following the success of deep learning in a wide range of applications, neural network-based machine-learning techniques have received significant interest for accelerating magnetic resonance imaging (MRI) acquisition and reconstruction strategies. A number of ideas inspired by deep learning techniques for computer vision and image processing have been successfully applied to nonlinear image reconstruction in the spirit of compressed sensing for accelerated MRI. Given the rapidly growing nature of the field, it is imperative to consolidate and summarize the large number of deep learning methods that have been reported in the literature, to obtain a better understanding of the field in general. This article provides an overview of the recent developments in neural-network based approaches that have been proposed specifically for improving parallel imaging. A general background and introduction to parallel MRI is also given from a classical view of k-space based reconstruction methods. Image domain based techniques that introduce improved regularizers are covered along with k-space based methods which focus on better interpolation strategies using neural networks. While the field is rapidly evolving with plenty of papers published each year, in this review, we attempt to cover broad categories of methods that have shown good performance on publicly available data sets. Limitations and open problems are also discussed and recent efforts for producing open data sets and benchmarks for the community are examined.

Scan-specific artifact reduction in k-space (SPARK) neural networks synergize with physics-based reconstruction to accelerate MRI

PURPOSE

To develop a scan-specific model that estimates and corrects k-space errors made when reconstructing accelerated MRI data.

METHODS

Scan-specific artifact reduction in k-space (SPARK) trains a convolutional-neural-network to estimate and correct k-space errors made by an input reconstruction technique by back-propagating from the mean-squared-error loss between an auto-calibration signal (ACS) and the input technique's reconstructed ACS. First, SPARK is applied to generalized autocalibrating partially parallel acquisitions (GRAPPA) and demonstrates improved robustness over other scan-specific models, such as robust artificial-neural-networks for k-space interpolation (RAKI) and residual-RAKI. Subsequent experiments demonstrate that SPARK synergizes with residual-RAKI to improve reconstruction performance. SPARK also improves reconstruction quality when applied to advanced acquisition and reconstruction techniques like 2D virtual coil (VC-) GRAPPA, 2D LORAKS, 3D GRAPPA without an integrated ACS region, and 2D/3D wave-encoded imaging.

RESULTS

SPARK yields SSIM improvement and 1.5 - 2× root mean squared error (RMSE) reduction when applied to GRAPPA and improves robustness to ACS size for various acceleration rates in comparison to other scan-specific techniques. When applied to advanced reconstruction techniques such as residual-RAKI, 2D VC-GRAPPA and LORAKS, SPARK achieves up to 20% RMSE improvement. SPARK with 3D GRAPPA also improves RMSE performance by ~2×, SSIM performance, and perceived image quality without a fully sampled ACS region. Finally, SPARK synergizes with non-Cartesian, 2D and 3D wave-encoding imaging by reducing RMSE between 20% and 25% and providing qualitative improvements.

CONCLUSION

SPARK synergizes with physics-based acquisition and reconstruction techniques to improve accelerated MRI by training scan-specific models to estimate and correct reconstruction errors in k-space.

End-to-end deep learning nonrigid motion-corrected reconstruction for highly accelerated free-breathing coronary MRA

PURPOSE

To develop an end-to-end deep learning technique for nonrigid motion-corrected (MoCo) reconstruction of ninefold undersampled free-breathing whole-heart coronary MRA (CMRA).

METHODS

A novel deep learning framework was developed consisting of a diffeomorphic registration network and a motion-informed model-based deep learning (MoDL) reconstruction network. The registration network receives as input highly undersampled (~22×) respiratory-resolved images and outputs 3D nonrigid respiratory motion fields between the images. The motion-informed MoDL performs MoCo reconstruction from undersampled data using the predicted motion fields. The whole deep learning framework, termed as MoCo-MoDL, was trained end-to-end in a supervised manner for simultaneous 3D nonrigid motion estimation and MoCo reconstruction. MoCo-MoDL was compared with a state-of-the-art nonrigid MoCo CMRA reconstruction technique in 15 retrospectively undersampled datasets and 9 prospectively undersampled acquisitions.

RESULTS

The acquisition time for ninefold accelerated CMRA was ~2.5 min. The reconstruction time was ~22 s for the proposed MoCo-MoDL and ~35 min for the conventional approach. MoCo-MoDL achieved higher peak SNR (27.86 ± 3.00 vs. 26.71 ± 2.79; P < .05) and structural similarity (0.78 ± 0.06 vs. 0.75 ± 0.06; P < .05) than the conventional approach. Similar vessel length and visual image quality score were obtained with the 2 methods, whereas improved vessel sharpness was observed with MoCo-MoDL.

CONCLUSION

An end-to-end deep learning approach was introduced for simultaneous nonrigid motion estimation and MoCo reconstruction of highly undersampled free-breathing whole-heart CMRA. The rapid free-breathing CMRA acquisition together with the fast reconstruction of the proposed approach promises easy integration into clinical workflow.

Feasibility and Implementation of a Deep Learning MR Reconstruction for TSE Sequences in Musculoskeletal Imaging

Magnetic Resonance Imaging (MRI) of the musculoskeletal system is one of the most common examinations in clinical routine. The application of Deep Learning (DL) reconstruction for MRI is increasingly gaining attention due to its potential to improve the image quality and reduce the acquisition time simultaneously. However, the technology has not yet been implemented in clinical routine for turbo spin echo (TSE) sequences in musculoskeletal imaging. The aim of this study was therefore to assess the technical feasibility and evaluate the image quality. Sixty examinations of knee, hip, ankle, shoulder, hand, and lumbar spine in healthy volunteers at 3 T were included in this prospective, internal-review-board-approved study. Conventional (TSE_S) and DL-based TSE sequences (TSE_DL) were compared regarding image quality, anatomical structures, and diagnostic confidence. Overall image quality was rated to be excellent, with a significant improvement in edge sharpness and reduced noise compared to TSE_S (p < 0.001). No difference was found concerning the extent of artifacts, the delineation of anatomical structures, and the diagnostic confidence comparing TSE_S and TSE_DL (p > 0.05). Therefore, DL image reconstruction for TSE sequences in MSK imaging is feasible, enabling a remarkable time saving (up to 75%), whilst maintaining excellent image quality and diagnostic confidence.

Dense Recurrent Neural Networks for Accelerated MRI: History-Cognizant Unrolling of Optimization Algorithms

Inverse problems for accelerated MRI typically incorporate domain-specific knowledge about the forward encoding operator in a regularized reconstruction framework. Recently physics-driven deep learning (DL) methods have been proposed to use neural networks for data-driven regularization. These methods unroll iterative optimization algorithms to solve the inverse problem objective function, by alternating between domain-specific data consistency and data-driven regularization via neural networks. The whole unrolled network is then trained end-to-end to learn the parameters of the network. Due to simplicity of data consistency updates with gradient descent steps, proximal gradient descent (PGD) is a common approach to unroll physics-driven DL reconstruction methods. However, PGD methods have slow convergence rates, necessitating a higher number of unrolled iterations, leading to memory issues in training and slower reconstruction times in testing. Inspired by efficient variants of PGD methods that use a history of the previous iterates, we propose a history-cognizant unrolling of the optimization algorithm with dense connections across iterations for improved performance. In our approach, the gradient descent steps are calculated at a trainable combination of the outputs of all the previous regularization units. We also apply this idea to unrolling variable splitting methods with quadratic relaxation. Our results in reconstruction of the fastMRI knee dataset show that the proposed history-cognizant approach reduces residual aliasing artifacts compared to its conventional unrolled counterpart without requiring extra computational power or increasing reconstruction time.

CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions

Cardiac CINE magnetic resonance imaging is the gold-standard for the assessment of cardiac function. Imaging accelerations have shown to enable 3D CINE with left ventricular (LV) coverage in a single breath-hold. However, 3D imaging remains limited to anisotropic resolution and long reconstruction times. Recently deep learning has shown promising results for computationally efficient reconstructions of highly accelerated 2D CINE imaging. In this work, we propose a novel 4D (3D + time) deep learning-based reconstruction network, termed 4D CINENet, for prospectively undersampled 3D Cartesian CINE imaging. CINENet is based on (3 + 1)D complex-valued spatio-temporal convolutions and multi-coil data processing. We trained and evaluated the proposed CINENet on in-house acquired 3D CINE data of 20 healthy subjects and 15 patients with suspected cardiovascular disease. The proposed CINENet network outperforms iterative reconstructions in visual image quality and contrast (+ 67% improvement). We found good agreement in LV function (bias ± 95% confidence) in terms of end-systolic volume (0 ± 3.3 ml), end-diastolic volume (− 0.4 ± 2.0 ml) and ejection fraction (0.1 ± 3.2%) compared to clinical gold-standard 2D CINE, enabling single breath-hold isotropic 3D CINE in less than 10 s scan and ~ 5 s reconstruction time.