Shu-Fu Shih^1,2, Sevgi Gokce Kafali^1,2, Kara L. Calkins³, and Holden H. Wu^1,2
1Department of Radiological Sciences, University of California Los Angeles, Los Angeles, CA, United States, ²Department of Bioengineering, University of California Los Angeles, Los Angeles, CA, United States, ³Department of Pediatrics, University of California Los Angeles, Los Angeles, CA, United States

MRI noninvasively quantifies liver fat and iron in terms of proton-density fat fraction (PDFF) and R₂*. While conventional Cartesian-based methods require breath-holding, recent self-gated free-breathing radial techniques have shown accurate and repeatable PDFF and R₂* mapping. However, data oversampling or computationally expensive reconstruction is required to reduce radial undersampling artifacts due to self-gating. This work developed an uncertainty-aware physics-driven deep learning network (UP-Net) that accurately and rapidly quantifies PDFF and R₂* using data from self-gated free-breathing stack-of-radial MRI. UP-Net used an MRI physics loss term to guide quantitative mapping, and also provided uncertainty estimation for each quantitative parameter.

Ivor J. A. Simpson¹, Ashley McManamon ¹, Alan J Stone², Nicholas P Blockley³, Alessandro Colasanti⁴, and Mara Cercignani^5
1Department of Informatics, University of Sussex, Brighton, United Kingdom, ²Department of Medical Physics and Clinical Engineering, St. Vincent's University Hospital, Dublin, Ireland, ³School of Life Sciences, University of Nottingham, Nottingham, United Kingdom, ⁴Brighton and Sussex Medical School, Brighton, United Kingdom, ⁵CUBRIC, Cardiff University, Cardiff, United Kingdom

Streamlined qBOLD acquisitions enable experimentally straightforward observations of brain metabolism. High quality R2’ maps are easily derived; however, the oxygen extraction fraction (OEF) and deoxygenated blood volume (DBV) are more ambiguously defined from noisy data. Accordingly, standard approaches yield noisy and underestimated OEF maps and overestimate DBV.

This work uses synthetic data to learn models for voxelwise prior distributions, which are subsequently leveraged in an amortized variational Bayesian inference model. We demonstrate our approach enables inference of smooth OEF and DBV maps, with a physiologically realistic distribution, and illustrate voxelwise differences in OEF between subjects at rest and undergoing hyperventilations. 

Jaejin Cho^1,2, Borjan Gagoski^2,3, Tae Hyung Kim^1,2, Qiyuan Tian^1,2, Robert Frost^1,2, Itthi Chatnuntawech⁴, and Berkin Bilgic^1,2,5
1Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA, United States, ²Department of Radiology, Harvard Medical School, Boston, MA, United States, ³Fetal-Neonatal Neuroimaging & Developmental Science Center, Boston Children’s Hospital, Boston, MA, United States, ⁴National Nanotechnology Center, Pathum Thani, Thailand, ⁵Harvard/MIT Health Sciences and Technology, Cambridge, MA, United States

We propose a wave-encoded model-based deep learning (wave-MoDL) method for joint multi-contrast image reconstruction with volumetric encoding using an interleaved look-locker acquisition sequence with T₂ preparation pulse (3D-QALAS). Wave-MoDL enables a 2-minute acquisition at R=4x3-fold acceleration using a 32-channel array to provide T₁, T₂, and proton density maps at 1 mm isotropic resolution, from which standard contrast-weighted images can also be synthesized. 

Chaowei Wu^1,2, Nan Wang³, Srinivas Gaddam⁴, Hui Han¹, Stephen Pandol⁴, Anthony G. Christodoulou^1,2, Yibin Xie¹, and Debiao Li^1,2
1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ²Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States, ³Radiology Department, Stanford University, Stanford, CA, United States, ⁴Division of Digestive and Liver Diseases, Cedars-Sinai Medical Center, Los Angeles, CA, United States

Quantitative dynamic contrast-enhanced (DCE) MRI has the potential for early detection, accurate staging, and therapy monitoring of cancers. However, clinical abdominal DCE-MRI has limited temporal resolution and can only provide qualitative or semi-quantitative assessments of tissue vascularity. In this study, we investigated the feasibility of retrospective quantification of multi-phasic abdominal DCE-MRI by improving the temporal resolution via deep learning. Simulated multi-phasic DCE data was generated using 2-sec temporal-resolution Multitasking DCE images. Results show that DCE kinetic parameters retrospectively estimated by deep learning agree with the ground truth, and are capable of differentiating abnormal tissues.

Ziyu Li¹, Qiuyun Fan^2,3, Berkin Bilgic^2,3, Guangzhi Wang⁴, Jonathan R Polimeni^2,3, Susie Y Huang^2,3, and Qiyuan Tian^2,3
1Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom, ²Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Boston, MA, United States, ³Harvard Medical School, Boston, MA, United States, ⁴Department of Biomedical Engineering, Tsinghua University, Beijing, China

The analysis of diffusion MRI data requires brain segmentation from separate anatomical images, which may be unavailable or cannot be accurately co-registered to diffusion images due to image distortions in diffusion data. Two state-of-the-art convolutional neural networks, U-Net and generative adversarial network (GAN), are employed to synthesize high-quality, distortion-matched T1w images directly from diffusion data, suitable for generating accurate cerebral cortical surfaces and volumetric segmentation for surface-based analysis of DTI metrics and tractography. The accuracy is quantitatively evaluated, and the systematical comparison shows that GAN-synthesized images are more visually appealing while U-Net-synthesized images achieve higher data consistency and segmentation accuracy.

Yujian Diao^1,2,3 and Ileana Ozana Jelescu^2,4
1Laboratory of Functional and Metabolic Imaging, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ²CIBM Center for Biomedical Imaging, Lausanne, Switzerland, ³Animal Imaging and Technology, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ⁴Department of Radiology, Lausanne University Hospital, Lausanne, Switzerland

WMTI-Watson is a widely used biophysical model that estimates microstructure parameters from the diffusion and kurtosis tensors. Here we propose a deep learning (DL) approach based on the recurrent neural network (RNN) to increase the robustness and accelerate the parameter estimation. The RNN solver achieved high accuracy, had good generality and was extremely fast in computation. The proposed DL approach is highly promising to replace the conventional nonlinear least-squares optimization in parameter estimation of WMTI-Watson model and thus estimate WM parameters from any DKI maps.