Jiyo S Athertya¹, Ya-Jun Ma¹, Amir Masoud Afsahi¹, Alicia Ji¹, Eric Y Chang^1,2, Jiang Du¹, and Hyungseok Jang^1
1Radiology, University of California San Diego, San Diego, CA, United States, ²Radiology Service, VA San Diego Healthcare System, San Diego, CA, United States

Quantitative ultrashort echo time (qUTE) imaging suffers from long acquisition time due to multiple acquisitions required for parameter estimation. In this study, feasibility of accelerated qUTE Cones imaging with compressed sensing (CS) reconstruction is investigated for fast variable flip angle UTE-T₁ mapping, adiabatic UTE-T_1ρ mapping, and UTE quantitative magnetization transfer (MT) modelling of macromolecular fraction (MMF). We explored these biomarkers for qUTE-Cones imaging of in vivo human knee joints at various undersampling rates. The performance of CS-reconstruction and parameter mapping was evaluated in tendons, ligaments, menisci, and cartilage.

Jiang Liu¹, Srivathsa Pasumarthi Venkata², and Keshav Datta^2
1Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, United States, ²R&D, Subtle Medical Inc, Menlo Park, CA, United States

Complementary information from multi-modal MRI is widely used in clinical practice for disease diagnosis. Due to scan time limitations, image corruptions, and different acquisition protocols, one or more contrasts may be missing or unusable. Recently developed CNN models for contrast synthesis are unable to capture the intricate dependencies between input contrasts and are not dynamic to the varying number of inputs. This work proposes a novel Multi-contrast and Multi-scale vision Transformer (MMT) that can take any number and combination of input sequences and synthesize the missing contrasts.

Alexander Jaffray^1,2, Zhe (Tim) Wu¹, Kamil Uludag^3,4, and Lars Kasper^1
1Techna Institute, University Health Network, Toronto, ON, Canada, ²UBC MRI Research Center, University of British Columbia, Vancouver, BC, Canada, ³Techna Institute & Koerner Scientist in MR Imaging, University Health Network, Toronto, ON, Canada, ⁴Biomedical Engineering, Center for Neuroscience Imaging Research, Institute for Basic Science, Sungkyunkwan University, Suwon, Korea, Republic of

We investigated the use of physically-motivated corrections for non-Cartesian imaging within a fast, open-source MR reconstruction framework written in the Julia programming language. Building on existing Julia libraries for MR image reconstruction, we developed a comprehensive, open-source pipeline for non-Cartesian image reconstruction including B₀ inhomogeneity correction, gradient system characterization and eddy current correction. We demonstrated that spiral images can be reconstructed using this pipeline with high geometric accuracy in a fast, accessible and efficient manner on modest hardware. 

D. Karkalousos¹, L. C. Liebrand¹, S. Noteboom², H. E. Hulst^2,3, F. M. Vos⁴, and M. W. A. Caan^1
1Department of Biomedical Engineering & Physics, Amsterdam UMC, University of Amsterdam, Amsterdam, Netherlands, ²Department of Anatomy & Neurosciences, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, Netherlands, ³Department of Medical, Health and Neuropsychology, Leiden University, Leiden, Netherlands, ⁴Delft University of Technology, Delft, Netherlands

Robustness when applying Deep Learning methods to clinical data is crucial for accurate high-resolution reconstructions while having fast inference times. We propose the Cascades of Independently Recurrent Variational Inference Machine (CIRVIM), targeting deep unrolled optimization and enforcing data consistency for further robustness. We quantify contrast resolution of seven and half times prospectively undersampled FLAIR MRI without fully-sampled center containing Multiple Sclerosis lesions. The proposed scheme reduces inference times by a factor of 6 compared to Compressed Sensing. Lesion contrast resolution improves by approximately 13% while preserving spatial detail with enhanced sharpness compared to more blurred results of other presented methods.

Laura Pfaff^1,2, Julian Hossbach^1,2, Elisabeth Preuhs¹, Tobias Wuerfl², Silvia Arroyo Camejo², Dominik Nickel², and Andreas Maier^1
1Pattern Recognition Lab, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany, ²MR Application Predevelopment, Siemens Healthcare GmbH, Erlangen, Germany

Accelerating MRI is intrinsically limited by the thermal noise from the imaged object. In this work we aim to optimize MR image denoising using an unsupervised deep learning-based method. Stein's unbiased risk estimator and spatially resolved noise maps indicating the standard deviation of the noise for every pixel were incorporated into the training process. It was shown that this approach can achieve results that are equal or superior to those of state-of-the-art supervised and unsupervised methods. Furthermore, we show how to control the tradeoff between denoising and image sharpness by using a model conditioned on the noise map.

Srivathsa Pasumarthi Venkata¹, Ben Andrew Duffy¹, Enhao Gong², Greg Zaharchuk³, and Keshav Datta^1
1R&D, Subtle Medical Inc, Menlo Park, CA, United States, ²R&D, Subtle Medical Inc., Menlo Park, CA, United States, ³Department of Radiology, Stanford University, Stanford, CA, United States

Complementary information from multi-contrast MRI data is used in deep learning algorithms for reducing contrast dosage in brain MRI. Though existing models produce clinically equivalent post-contrast images, they lack explainability in terms of mapping the source of contrast information from input to output. In this work we explore the feasibility of an explainable deep learning model for gadolinium dose reduction in contrast-enhanced brain MRI.

Nicholas Dwork¹, Shuyu Tang¹, Peder E. Z. Larson¹, and Jeremy W. Gordon^1
1Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States

We present a least-squares method for estimating sensitivity when imaging x-nuclei.  We show results with hyperpolarized pyruvate in both the heart and the brain.

Parna Eshraghi Boroojeni¹, Paul Commean¹, Cihat Eldeniz¹, Weijie Gan¹, Gary Skolnick¹, Kamlesh Patel¹, Ulugbek Kamilov¹, and Hongyu An^1
1Washington University in Saint Louis, Saint Louis, MO, United States

A high-resolution (HR) MRI capable of resolving the detail of bony structures at sub-millimeter resolution is desired. A short MR acquisition results in under-sampled k-space data below the Nyquist rate, leading to artifacts and high noise. We developed an HR reconstruction method regularized by a complex deep-learning prior (RECD). We achieved high-resolution MR (0.6x0.6x0.8mm3) with a one-minute acquisition time. Using images reconstructed from a 5-minute MR scan as the gold standard, we compared the peak signal to noise ratio (PSNR) and similarity index (SSIM) for 1-min RECD and 1-min compressed sensing (CS) reconstructed images. RECD outperformed CS.

Gabriel Varela-Mattatall^1,2, Roy A.M. Haast ^1,3, Omer Oran⁴, Ali R. Khan ^1,2, and Ravi S. Menon^1,2
1Centre for Functional and Metabolic Mapping (CFMM) | Robarts Research Institute | Western University, London, ON, Canada, ²Department of Medical Biophysics | Schulich School of Medicine and Dentistry | Western University, London, ON, Canada, ³Aix-Marseille Universite | CNRS | CRMBM, Marseille, France, ⁴Siemens Healthcare Limited, Oakville, ON, Canada

With the recent development of an automatic compressed sensing (CS)-based reconstruction method, we investigated its allowance to recover T₂^* maps from prospectively undersampled multi-echo, gradient echo-based acquisitions at 7T. We compared our CS method with the more conventional GRAPPA-based reconstruction using up to 16-fold acceleration. In contrast to GRAPPA, our CS-based method allows recovery of T₂^* maps up to 16-fold acceleration, which demonstrates its promise for ultra-fast acquisitions. However, current results also show some caveats that need to be addressed as future work.

Aziz Kocanaogullari¹, Cemre Ariyurek¹, Onur Afacan¹, and Sila Kurugol^1
1Radiology, Boston Children's Hospital, Boston, MA, United States

MRI literature shows that it is possible to increase image reconstruction quality by removing coils that cause majority of the streaking artifacts. We introduce to measure the coil quality based on mutual information between a reference and each coil image. Specifically we apply this technique to DCE-MRI reconstruction from under-sampled radial stack of stars trajectory for kidney imaging and calculate mutual dependence between a dynamic image from each coil and a reference image to assess coil contribution to reconstruction. Experiments show mutual dependence based coil selection reduces artifacts and increases reconstructed image SNR by $$$16.84\%$$$ using $$$1/3$$$ of the coils.

Jaeyong Yu^1,2, Sugil Kim³, and Jang-Yeon Park^1,2
1Department of Biomedical Engineering, Sungkyunkwan University, Suwon, Korea, Republic of, ²Department of Intelligent Precision Healthcare Convergence, Sungkyunkwan University, Suwon, Korea, Republic of, ³Siemens Healthineers Ltd., Seoul, Korea, Republic of

To reduce acquisition time, acceleration techniques such as simultaneous multi-slice and parallel imaging techniques have been applied to various sequences. In this work, we proposed a reconstruction method for simultaneous multi-slab spatiotemporal-encoding (SMS SPEN) using split slice-GRAPPA combined with CAIPIRINHA. The proposed method was applied to the recently developed ultrafast 3D SPEN imaging sequence called ERASE (Equal-TE Rapid Acquisition with Sequential Excitation) and a case of multiband factor of 3 was demonstrated with human brain imaging at 3T.