Deep variational network for rapid 4D flow MRI reconstruction

Vishnevskiy, Valeriy; Walheim, Jonas; Kozerke, Sebastian

doi:10.1038/s42256-020-0165-6

Cited by 59 publications

(52 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…15 Deep learning (DL) -based aliasing artifact removal in accelerated cardiovascular MRI has been proposed as an alternative to CS to reduce total reconstruction time [16][17][18][19][20] and improve performance. [19][20][21] Moreover, recent studies have shown the utility of DL-based methods for PC MRI. Vishnevskiy et al 21 showed that an unrolled network incorporating a physics-based model into the DL architecture reduces reconstruction time 30-fold compared to CS for 12.4-to 13.8fold accelerated 4D ECG-gated segmented PC MRI with 25 cardiac phases.…”

Section: Introductionmentioning

confidence: 99%

“…[19][20][21] Moreover, recent studies have shown the utility of DL-based methods for PC MRI. Vishnevskiy et al 21 showed that an unrolled network incorporating a physics-based model into the DL architecture reduces reconstruction time 30-fold compared to CS for 12.4-to 13.8fold accelerated 4D ECG-gated segmented PC MRI with 25 cardiac phases. Ferdian et al 22 showed that 4DFlowNet network trained using synthetic 4D flow MR generated from computational fluid dynamic (CFD) solutions could be used to increase spatial resolution.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Highly accelerated free‐breathing real‐time phase contrast cardiovascular MRI via complex‐difference deep learning

Haji‐Valizadeh

Guo

Küçükseymen

et al. 2021

Magnetic Resonance in Med

View full text Add to dashboard Cite

Purpose: To develop and evaluate a real-time phase contrast (PC) MRI protocol via complex-difference deep learning (DL) framework. Methods: DL used two 3D U-nets to separately filter aliasing artifact from radial real-time velocity-compensated and complex-difference images. U-nets were trained with synthetic real-time PC generated from electrocardiograph (ECG) -gated, breathhold, segmented PC (ECG-gated segmented PC) acquired at the ascending aorta of 510 patients. In 21 patients, free-breathing, ungated real-time (acceleration rate = 28.8) and ECG-gated segmented (acceleration rate = 2) PC were prospectively acquired at the ascending aorta. Hemodynamic parameters (cardiac output [CO], stroke volume [SV], and mean velocity at peak systole [peak mean velocity]) were measured for ECG-gated segmented and DL-filtered synthetic real-time PC and compared using Bland-Altman and linear regression analyses. Additionally, hemodynamic parameters were quantified from DL-filtered, compressed-sensing (CS) -reconstructed, and gridding reconstructed prospective real-time PC and compared to ECG-gated segmented PC. Results: Synthetic real-time PC with DL showed strong correlation (R > 0.98) and good agreement with ECG-gated segmented PC for quantified hemodynamic parameters (mean-difference: CO = −0.3 L/min, SV = −4.3 mL, peak mean velocity = −2.3 cm/s). On average, DL required 0.39 s/frame to filter prospective real-time PC, which was 4.6-fold faster than CS. Compared to CS, DL showed superior correlation, tighter limits of agreement (LOAs), better bias for peak mean velocity, and worse bias for CO and SV. Compared to gridding, DL showed similar correlation, tighter LOAs for CO and SV, similar bias for CO, and worse bias for SV and peak mean velocity. Conclusion:The complex-difference DL framework accelerated real-time PC-MRI by nearly 28-fold, enabling rapid free-running real-time assessment of flow hemodynamics.How to cite this article: Haji-Valizadeh H, Guo R, Kucukseymen S, et al. Highly accelerated free-breathing real-time phase contrast cardiovascular MRI via complex-difference deep learning.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Highly accelerated free‐breathing real‐time phase contrast cardiovascular MRI via complex‐difference deep learning

Haji‐Valizadeh

Guo

Küçükseymen

et al. 2021

Magnetic Resonance in Med

View full text Add to dashboard Cite

show abstract

“…In fact, a subsequent step is still required to translate the new methods proposed here and based on subvoxel modelling to the clinical practice. In fact, given the recent advances in the 4D Flow magnetic resonance technique to assess ensembleaveraged velocities as well as the full Reynolds stress tensor [30], [31], it is foreseeable that the approach presented in this study will be applicable once a validated sequence for the evaluation of the first two terms of Eq. 19 and the ensembleaveraged cubed magnitude of the instantaneous rate-of-strain tensor will be available.…”

Section: Limitationsmentioning

confidence: 99%

A Novel Estimation Approach of Pressure Gradient and Haemodynamic Stresses as Indicators of Pathological Aortic Flow Using Subvoxel Modelling

Corso¹,

Giannakopoulos²,

Gülan³

et al. 2021

IEEE Trans. Biomed. Eng.

View full text Add to dashboard Cite

The flow downstream from aortic stenoses 1 is characterised by the onset of shear-induced turbulence that 2 leads to irreversible pressure losses. These extra losses represent 3 an increased resistance that impacts cardiac efficiency. A novel 4 approach is suggested in this study to accurately evaluate 5 the pressure gradient profile along the aorta centreline using 6 modelling of haemodynamic stress at scales that are smaller 7 than the typical resolution achieved in experiments. Methods: We 8 use benchmark data obtained from direct numerical simulation 9 (DNS) along with results from in silico and in vitro three-10 dimensional particle tracking velocimetry (3D-PTV) at three 11 voxel sizes, namely 750 µm, 1 mm and 1.5 mm. A differential 12 equation is derived for the pressure gradient, and the subvoxel-13 scale (SVS) stresses are closed using the Smagorinsky and a 14 new refined model. Model constants are optimised using DNS 15 and in silico PTV data and validated based on pulsatile in 16 vitro 3D-PTV data and pressure catheter measurements. Results: 17 The Smagorinsky-based model was found to be more accurate 18 for SVS stress estimation but also more sensitive to errors 19 especially at lower resolution, whereas the new model was found 20 to more accurately estimate the projected pressure gradient even 21 for larger voxel size of 1.5mm albeit at the cost of increased 22 sensitivity at this voxel size. A comparison with other methods 23 in the literature shows that the new approach applied to in 24 vitro PTV measurements estimates the irreversible pressure 25 drop by decreasing the errors by at least 20%. Conclusion: 26 Our novel approach based on the modelling of subvoxel stress 27 offers a validated and more accurate way to estimate pressure 28 gradient, irreversible pressure loss and SVS stress. Significance: 29 We anticipate that the approach may potentially be applied to 30 image-based in vivo, in vitro 4D flow data or in silico data with 31 limited spatial resolution to assess pressure loss and SVS stresses 32 in disturbed aortic blood flow.

show abstract

“…The use of high-performance computing (HPC) is gaining ground in high-dimensional imaging data processing [16], as in the context of hyperspectral image processing [5,35] and medical image analysis [12]. In particular, for the specific case of medical imaging, along with the acceleration of the training of deep neural networks [47], graphics processing unit (GPU)-powered implementations allowed for real-time performance in image reconstruction [46,59], segmentation [2], as well as feature extraction [42] and classification [22]. Moreover, multi-core and many-core architectures were exploited to accelerate computationally expensive medical image enhancement and quantification tasks [41,52,53].…”

Section: Introductionmentioning

confidence: 99%

A CUDA-powered method for the feature extraction and unsupervised analysis of medical images

et al. 2021

View full text Add to dashboard Cite

Image texture extraction and analysis are fundamental steps in computer vision. In particular, considering the biomedical field, quantitative imaging methods are increasingly gaining importance because they convey scientifically and clinically relevant information for prediction, prognosis, and treatment response assessment. In this context, radiomic approaches are fostering large-scale studies that can have a significant impact in the clinical practice. In this work, we present a novel method, called CHASM (Cuda, HAralick & SoM), which is accelerated on the graphics processing unit (GPU) for quantitative imaging analyses based on Haralick features and on the self-organizing map (SOM). The Haralick features extraction step relies upon the gray-level co-occurrence matrix, which is computationally burdensome on medical images characterized by a high bit depth. The downstream analyses exploit the SOM with the goal of identifying the underlying clusters of pixels in an unsupervised manner. CHASM is conceived to leverage the parallel computation capabilities of modern GPUs. Analyzing ovarian cancer computed tomography images, CHASM achieved up to $$\sim 19.5\times $$ ∼ 19.5 × and $$\sim 37\times $$ ∼ 37 × speed-up factors for the Haralick feature extraction and for the SOM execution, respectively, compared to the corresponding C++ coded sequential versions. Such computational results point out the potential of GPUs in the clinical research.

show abstract

Deep variational network for rapid 4D flow MRI reconstruction

Cited by 59 publications

References 40 publications

Highly accelerated free‐breathing real‐time phase contrast cardiovascular MRI via complex‐difference deep learning

Highly accelerated free‐breathing real‐time phase contrast cardiovascular MRI via complex‐difference deep learning

A Novel Estimation Approach of Pressure Gradient and Haemodynamic Stresses as Indicators of Pathological Aortic Flow Using Subvoxel Modelling

A CUDA-powered method for the feature extraction and unsupervised analysis of medical images

Contact Info

Product

Resources

About