GRASPNET: Fast spatiotemporal deep learning reconstruction of golden‐angle radial data for free‐breathing dynamic contrast‐enhanced magnetic resonance imaging

Jafari, Ramin; Gian, Richard Kinh; LaGratta, Maria D.; Fung, Maggie; Bayram, Ersin; Cashen, Ty A.; Otazo, Ricardo

doi:10.1002/nbm.4861

Cited by 9 publications

(8 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The development of rapid radial reconstruction techniques is of interest for MRIgRT, where the latencies associated with iterative reconstruction are prohibitively long. 46 While our proof -of -principle study has focused on the application of AUTOMAP to real-time targeting of radiotherapy in the lung, we believe our results are extensible to high-motion sites such as the liver and prostate, where tumor movement would be optimally managed by real-time adaptive radiotherapy. 47,48 We note that due to the relatively high latency of MR acquisition and reconstruction, compared to X-ray-based modalities, faster image reconstruction techniques are desired for real-time beam gating and MLC tracking on MRI-Linacs, especially for non-Cartesian acquisition trajectories.…”

Section: Discussionmentioning

confidence: 85%

“…Our results leverage advances in machine learning to implement fast image reconstruction of undersampled radial data from lung cancer patients with comparable accuracy to conventional iterative reconstruction techniques based on CS. The development of rapid radial reconstruction techniques is of interest for MRIgRT, where the latencies associated with iterative reconstruction are prohibitively long 46 . While our proof‐of‐principle study has focused on the application of AUTOMAP to real‐time targeting of radiotherapy in the lung, we believe our results are extensible to high‐motion sites such as the liver and prostate, where tumor movement would be optimally managed by real‐time adaptive radiotherapy 47,48 .…”

Section: Discussionmentioning

confidence: 89%

See 1 more Smart Citation

Real‐time radial reconstruction with domain transform manifold learning for MRI‐guided radiotherapy

et al. 2023

View full text Add to dashboard Cite

Background: MRI-guidance techniques that dynamically adapt radiation beams to follow tumor motion in real time will lead to more accurate cancer treatments and reduced collateral healthy tissue damage. The gold-standard for reconstruction of undersampled MR data is compressed sensing (CS) which is computationally slow and limits the rate that images can be available for real-time adaptation. Purpose: Once trained, neural networks can be used to accurately reconstruct raw MRI data with minimal latency. Here, we test the suitability of deep-learningbased image reconstruction for real-time tracking applications on MRI-Linacs. Methods: We use automated transform by manifold approximation (AUTOMAP), a generalized framework that maps raw MR signal to the target image domain, to rapidly reconstruct images from undersampled radial k-space data. The AUTOMAP neural network was trained to reconstruct images from a golden-angle radial acquisition, a benchmark for motion-sensitive imaging, on lung cancer patient data and generic images from ImageNet. Model training was subsequently augmented with motion-encoded k-space data derived from videos in the YouTube-8M dataset to encourage motion robust reconstruction. Results: AUTOMAP models fine-tuned on retrospectively acquired lung cancer patient data reconstructed radial k-space with equivalent accuracy to CS but with much shorter processing times. Validation of motion-trained models with a virtual dynamic lung tumor phantom showed that the generalized motion properties learned from YouTube lead to improved target tracking accuracy. Conclusion: AUTOMAP can achieve real-time, accurate reconstruction of radial data. These findings imply that neural-network-based reconstruction is potentially superior to alternative approaches for real-time image guidance applications.

show abstract

Section: Discussionmentioning

confidence: 85%

Section: Discussionmentioning

confidence: 89%

Real‐time radial reconstruction with domain transform manifold learning for MRI‐guided radiotherapy

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Reconstruction of the motion dictionary was implemented in MATLAB to show feasibility of MRSIGMA, which is very slow for clinical implementation. Future work will explore the use of deep learning to further reduce the scan time and most significantly reduce the reconstruction time of the motion dictionary from hours to seconds, as recently demonstrated by our group (Jafari et al 2021). Second, there were only 5 patients in this initial feasibility study, which are not enough to evaluate statistical significance.…”

Section: Discussionmentioning

confidence: 92%

MR SIGnature MAtching (MRSIGMA) with retrospective self-evaluation for real-time volumetric motion imaging

et al. 2021

View full text Add to dashboard Cite

Objective. MR SIGnature MAtching (MRSIGMA) is a real-time volumetric MRI technique to image tumor and organs at risk motion in real-time for radiotherapy applications, where a dictionary of high-resolution 3D motion states and associated motion signatures are computed first during offline training and real-time 3D imaging is performed afterwards using fast signature-only acquisition and signature matching. However, the lack of a reference image with similar spatial resolution and temporal resolution introduces significant challenges for in vivo validation. Approach. This work proposes a retrospective self-validation for MRSIGMA, where the same data used for real-time imaging are used to create a non-real-time reference for comparison. MRSIGMA with self-validation is tested in patients with liver tumors using quantitative metrics defined on the tumor and nearby organs-at-risk structures. The dice coefficient between contours defined on the real-time MRSIGMA and non-real-time reference was used to assess motion imaging performance. Main Results. Total latency (including signature acquisition and signature matching) was between 250 and 314 ms, which is sufficient for organs affected by respiratory motion. Mean ± standard deviation dice coefficient over time was 0.74 ± 0.03 for patients imaged without contrast agent and 0.87 ± 0.03 for patients imaged with contrast agent, which demonstrated high-performance real-time motion imaging. Signficance. MRSIGMA with self-evaluation provides a means to perform real-time volumetric MRI for organ motion tracking with quantitative performance measures.

show abstract

“…This concept is not limited to the case of motion‐resolved imaging and can be used for other cases with dynamic information, such as contrast‐enhanced imaging, in which the input image to the network can also be sorted into a particular order of contrast phases. In fact, previous work on dynamic contrast‐enhanced MRI implicitly exploited a similar property to remove data consistency 29 …”

Section: Discussionmentioning

confidence: 99%

“…In fact, previous work on dynamic contrast-enhanced MRI implicitly exploited a similar property to remove data consistency. 29 A comparison of Movienet to unrolled reconstruction networks was not explicitly performed in this work. Instead, Movienet was compared against XD-GRASP, which is a state-of-the-art technique for motion-resolved 4D MRI using golden-angle radial sampling that uses explicit k-space data consistency.…”

Section: Discussionmentioning

confidence: 99%

Movienet: Deep space–time‐coil reconstruction network without k‐space data consistency for fast motion‐resolved 4D MRI

Murray,

Siddiq,

Crane

et al. 2023

Magnetic Resonance in Med

Self Cite

View full text Add to dashboard Cite

PurposeTo develop a novel deep learning approach for 4D‐MRI reconstruction, named Movienet, which exploits space–time‐coil correlations and motion preservation instead of k‐space data consistency, to accelerate the acquisition of golden‐angle radial data and enable subsecond reconstruction times in dynamic MRI.MethodsMovienet uses a U‐net architecture with modified residual learning blocks that operate entirely in the image domain to remove aliasing artifacts and reconstruct an unaliased motion‐resolved 4D image. Motion preservation is enforced by sorting the input image and reference for training in a linear motion order from expiration to inspiration. The input image was collected with a lower scan time than the reference XD‐GRASP image used for training. Movienet is demonstrated for motion‐resolved 4D MRI and motion‐resistant 3D MRI of abdominal tumors on a therapeutic 1.5T MR‐Linac (1.5‐fold acquisition acceleration) and diagnostic 3T MRI scanners (2‐fold and 2.25‐fold acquisition acceleration for 4D and 3D, respectively). Image quality was evaluated quantitatively and qualitatively by expert clinical readers.ResultsThe reconstruction time of Movienet was 0.69 s (4 motion states) and 0.75 s (10 motion states), which is substantially lower than iterative XD‐GRASP and unrolled reconstruction networks. Movienet enables faster acquisition than XD‐GRASP with similar overall image quality and improved suppression of streaking artifacts.ConclusionMovienet accelerates data acquisition with respect to compressed sensing and reconstructs 4D images in less than 1 s, which would enable an efficient implementation of 4D MRI in a clinical setting for fast motion‐resistant 3D anatomical imaging or motion‐resolved 4D imaging.

show abstract

GRASPNET: Fast spatiotemporal deep learning reconstruction of golden‐angle radial data for free‐breathing dynamic contrast‐enhanced magnetic resonance imaging

Cited by 9 publications

References 31 publications

Real‐time radial reconstruction with domain transform manifold learning for MRI‐guided radiotherapy

Real‐time radial reconstruction with domain transform manifold learning for MRI‐guided radiotherapy

MR SIGnature MAtching (MRSIGMA) with retrospective self-evaluation for real-time volumetric motion imaging

Movienet: Deep space–time‐coil reconstruction network without k‐space data consistency for fast motion‐resolved 4D MRI

Contact Info

Product

Resources

About