Model-based deep learning reconstruction methods for fast magnetic resonance imaging

다국어 초록 (Multilingual Abstract)

Magnetic resonance imaging (MRI) provides visualization inside the human body noninvasively and is a widely used medical imaging modality for the diagnosis of patients. Compared to other medical imaging modalities magnetic resonance (MR) images present high spatial resolution and high soft-tissue contrast, giving valuable diagnostic information. Nonetheless, MRI takes a long scan time for acquiring raw MR data in the k-space domain, which leads to motion artifacts due to patients’ discomfort and hinders the further applications of MRI. For acceleration of MRI, undersampling the k-space data can be conducted (i.e., sampling below the Nyquist rate), but it evokes aliasing artifacts in the reconstructed image. In this dissertation, three different deep learning-based reconstruction methods are introduced for fast MRI. The first method is a deep parallel imaging network (“DPI-net”) that reconstructs 3D time-of-flight magnetic resonance angiography from undersampled multi-coil k-space data using multi-stream convolutional neural networks. The second method is a joint deep model-based MR image and coil sensitivity reconstruction network (“Joint-ICNet”) that jointly reconstructs an MR image and coil sensitivity maps from undersampled multi-coil k-space data using deep learning networks and interleaved MR model-based data consistency schemes. The third method is a deep model-based MR parameter mapping network (“DOPAMINE”) that reconstructs MR parameter maps from undersampled multi-coil k-space data using deep learning networks combined with MR physical models. DPI-net, Joint-ICNet, and DOPAMINE outperformed conventional MR image reconstruction methods such as parallel imaging and compressive sensing-based methods and state-of-the-art deep learning-based methods in reconstructing MR image and MR parameter maps from undersampled multi-coil k-space data. Therefore, the proposed deep learning-based reconstruction methods can be applied to the acceleration of MRI.

국문 초록 (Abstract)

자기공명영상은 비침습적으로 인체 내부를 시각화하여 환자 진단에 널리 사용되는 의료 영상 기법이다. 다른 의료 영상 기법과 비교하여 자기공명 영상은 높은 공간 해상도와 연조직 대비를 나타내어 중요한 진단 정보를 제공한다. 그러나, 자기공명영상은 k-space 영역에서 원 자기공명 데이터를 획득하는 데 긴 촬영 시간이 소요되며, 이는 환자의 불편함으로 인한 움직임 인공물을 유발하고 자기공명영상의 추가 적용을 저해한다. 자기공명영상의 가속화를 위해 k-space 전체 데이터를 획득하는 대신 나이퀴스트 샘플링 레이트 아래로 과소 샘플링 될 수 있지만 공간 영역에서 에일리어싱 인공물을 유발한다. 본 논문에서는 고속 자기공명영상을 위한 세 가지 딥러닝 기반 재구성 방법론을 소개한다. 첫 번째 방법론은 다중 스트림 컨볼루션 신경망을 사용하여 과소샘플링된 다중 코일 k-space 데이터로부터 3D time-of-flight 자기공명혈관조영술을 재구성하는 심층 병렬 이미징 네트워크(“DPI-net”)이다. 두 번째 방법론은 딥러닝 네트워크와 삽입된 자기공명 모델 기반 데이터 유지 방법을 사용하여 과소 샘플링된 다중 코일 k-space 데이터에서 자기공명 영상과 코일 감도 맵을 같이 재구성하는 조인트 심층 모델 기반 자기공명 영상 및 코일 감도 재구성 네트워크(“Joint-ICNet”)이다. 세 번째 방법론은 자기공명 물리 모델과 결합된 딥러닝 네트워크를 사용하여 과소 샘플링된 다중 코일 k-space 데이터에서 자기공명 파라미터 맵을 재구성하는 심층 모델 기반 자기공명 파라미터 매핑 네트워크(“DOPAMINE”)이다. DPI-net, Joint-ICNet, 및 DOPAMINE은 과소 샘플링된 다중 코일 k-space 데이터에서 자기공명 영상 및 자기공명 파라미터 맵을 재구성하는 데 있어 병렬 이미징 및 압축 센싱 기반 방법과 같은 기존의 자기공명 영상 재구성 방법론들과 최신 딥러닝 기반 방법론들을 능가하였다. 따라서, 제안된 딥러닝 기반 재구성 방법론은 자기공명영상 고속화에 적용될 수 있다.

참고문헌 (Reference)

1 Lustig , M., "Compressed sensing MRI .", 25 ( 2 ) : p. 72-82 ., 2008

2 Graves , M. ,, "Magnetic resonance angiography", 70 ( 829 ) : p. 6-28 ., 1997

3 He , K. , et al, "Deep residual learning for image recognition", 2016

4 Song , P., "HYDRA : Hybrid deep magnetic resonance fingerprinting", 46 ( 11 ) : p. 4951-4969 ., 2019

5 Zhang , K. , et al, "Learning deep CNN denoiser prior for image restoration", 2017

6 Zhu , B. ,, "Image reconstruction by domain-transform manifold learning", 555 ( 7697 ) : p. 487-492 ., 2018

7 Knoll , F., "Second order total generalized variation ( TGV ) for MRI .", 65 ( 2 ) : p. 480-491 ., 2011

8 Xie , S. , et al, "Aggregated residual transformations for deep neural networks", 2017

9 Stikov , N., "On the accuracy of T1 mapping : searching for common ground .", 73 ( 2 ) : p. 514-522 ., 2015

10 Liu , G. , et al ., "Image inpainting for irregular holes using partial convolutions .", 2018

11 Zhang , T., "Coil compression for accelerated imaging with Cartesian sampling .", 69 ( 2 ) : p. 571-582 ., 2013

12 Lim , B. , et al, "Enhanced deep residual networks for single image superresolution .", 2017

13 Jung , W. ,, "Exploring linearity of deep neural network trained QSM : QSMnet+ .", 211 : p. 116619 ., 2020

14 Yang , Y., "ADMM-CSNet : A deep learning approach for image compressive sensing .", 42 ( 3 ) : p. 521-538 ., 2018

15 Griswold , M.A. ,, "Generalized autocalibrating partially parallel acquisitions ( GRAPPA )", 47 ( 6 ) : p. 1202-1210 ., 2002

16 Yoon , J. ,, "Quantitative susceptibility mapping using deep neural network : QSMnet", 179 : p. 199-206 ., 2018

17 Mardani , M., "Deep generative adversarial neural networks for compressive sensing MRI .", 38 ( 1 ) : p. 167-179 ., 2018

18 Zhao , B., "Accelerated MR parameter mapping with low ? rank and sparsity constraints .", 74 ( 2 ) : p. 489- 498 ., 2015

19 Wang , Z., "Image quality assessment : from error visibility to structural similarity .", 13 ( 4 ) : p. 600-612 ., 2004

20 Hammernik , K., "Learning a variational network for reconstruction of accelerated MRI data .", 79 ( 6 ) : p. 3055-3071 ., 2018

21 Lee , D., "Deep residual learning for accelerated MRI using magnitude and phase networks .", 65 ( 9 ) : p. 1985-1995 ., 2018

22 Zhang , K., "Beyond a gaussian denoiser : Residual learning of deep cnn for image denoising .", 26 ( 7 ) : p. 3142-3155 ., 2017

23 Duan , J. , et al ., "VS-Net : Variable splitting network for accelerated parallel MRI reconstruction .", 2019

24 Schlemper , J., "A deep cascade of convolutional neural networks for dynamic MR image reconstruction .", 37 ( 2 ) : p. 491-503 ., 2017

25 Keenan , K.E., "Recommendations towards standards for quantitative MRI ( qMRI ) and outstanding needs .", 49 ( 7 ) : p. e26 ., 2019

26 Maier , O., "Rapid T1 quantification from high resolution 3D data with model ? based reconstruction .", 81 ( 3 ) : p. 2072-2089 ., 2019

27 K ? stner , T., "Retrospective correction of motion ? affected MR images using deep learning frameworks .", 82 ( 4 ) : p. 1527-1540 ., 2019

28 He , K. , et al ., "Delving deep into rectifiers : Surpassing human-level performance on imagenet classification", 2015

29 Cheng , D. , et al, "Person re-identification by multi-channel parts-based cnn with improved triplet loss function", 2016

30 Eo , T., "Accelerating Cartesian MRI by domain-transform manifold learning in phase-encoding direction .", 63 : p. 101689 ., 2020

31 Shin , P.J., "Calibrationless parallel imaging reconstruction based on structured low ? rank matrix completion", 72 ( 4 ) : p. 959-970 ., 2014

32 Chavhan , G.B., "Principles , techniques , and applications of T2 * -based MR imaging and its special applications", 29 ( 5 ) : p. 1433-1449 ., 2009

33 Gibbs , G.F. ,, "Improved image quality of intracranial aneurysms : 3.0- T versus 1.5-T time-of-flight MR angiography", 25 ( 1 ) : p. 84-87 ., 2004

34 Yang , G., "DAGAN : Deep de-aliasing generative adversarial networks for fast compressed sensing MRI reconstruction .", 37 ( 6 ) : p. 1310-1321 ., 2017

35 Baudrexel , S., "Quantitative mapping of T1 and T2 * discloses nigral and brainstem pathology in early Parkinson 's disease", 51 ( 2 ) : p. 512-520 ., 2010

36 Eo , T. KIKI ?, "net : cross ? domain convolutional neural networks for reconstructing undersampled magnetic resonance images .", 80 ( 5 ) : p. 2188-2201 ., 2018

37 Oelerich , M. ,, "Intracranial vascular stenosis and occlusion : comparison of 3D time-of-flight and 3D phase-contrast MR angiography", 40 ( 9 ) : p. 567-573 ., 1998

38 Uecker , M., "Image reconstruction by regularized nonlinear inversion ? joint estimation of coil sensitivities and image content .", 60 ( 3 ) : p. 674-682 ., 2008

39 Murphy , M. ,, "Fast l1-SPIRiT compressed sensing parallel imaging MRI : scalable parallel implementation and clinically feasible runtime", 31 ( 6 ) : p. 1250-1262 ., 2012

40 Jun , Y., "Parallel imaging in time ? of ? flight magnetic resonance angiography using deep multistream convolutional neural networks .", 81 ( 6 ) : p. 3840-3853 ., 2019

41 Huang , C., "T2 mapping from highly undersampled data by reconstruction of principal component coefficient maps using compressed sensing .", 67 ( 5 ) : p. 1355-1366 ., 2012

42 Sumpf , T.J., "Fast T2 mapping with improved accuracy using undersampled spin-echo MRI and model-based reconstructions with a generating function .", 33 ( 12 ) : p. 2213-2222 ., 2014

43 Fushimi , Y., "Compressed sensing 3-dimensional time-of-flight magnetic resonance angiography for cerebral aneurysms : optimization and evaluation .", 51 ( 4 ) : p. 228-235 ., 2016

44 Weber , J. ,, "MR angiography at 3 Tesla to assess proximal internal carotid artery stenoses : contrast-enhanced or 3D time-of-flight MR angiography ?", 25 ( 1 ) : p. 41-48 ., 2015

45 Zhang , T., "Clinical performance of contrast enhanced abdominal pediatric MRI with fast combined parallel imaging compressed sensing reconstruction .", 40 ( 1 ) : p. 13- 25 ., 2014

46 Wang , X., "Model-based myocardial T1 mapping with sparsity constraints using single-shot inversion-recovery radial FLASH cardiovascular magnetic resonance .", 21 ( 1 ) : p. 1-11 ., 2019

47 Lescher , S., "Quantitative T1 and T2 mapping in recurrent glioblastomas under bevacizumab : earlier detection of tumor progression compared to conventional MRI .", 57 ( 1 ) : p. 11-20 ., 2015

48 Yamamoto , T., "Time-of-flight magnetic resonance angiography with sparse undersampling and iterative reconstruction : comparison with conventional parallel imaging for accelerated imaging", 51 ( 6 ) : p. 372-378 ., 2016

49 Kramer , J.H. ,, "Dynamic and static magnetic resonance angiography of the supra-aortic vessels at 3.0 T : intraindividual comparison of gadobutrol , gadobenate dimeglumine , and gadoterate meglumine at equimolar dose .", 48 ( 3 ) : p. 121-128 ., 2013