Chenyu He¹, Xiaojun Guan², Boliang Yu¹, Hongjiang Wei³, and Yuyao Zhang^1,4,5
1School of Information Science and Technology, ShanghaiTech University, Shanghai, China, ²Department of Radiology, The Second Affiliated Hospital, Zhejiang, University School of Medicine, Hangzhou, China, ³Institute for Medical Imaging Technology, School of Biomedical Engineering, Shanghai, Jiao Tong University, Shanghai, China, ⁴Shanghai Engineering Research Center of Intelligent Vision and Imaging, ShanghaiTech University, Shanghai, China, ⁵iHuman Institute, ShanghaiTech University, Shanghai, China

Our work successfully provides precious manual depiction of typical DBNs in the standard MNI space with the guidance of constructed MNI-spaced average multi-modal template (T1-weighted and quantitative susceptibility mapping (QSM)). 10 pairs of DBN and their sub-nuclei, are manually depicted including caudate nucleus (CN), putamen (Pu), ventral pallidum (VP), globus pallidus (GP) (internal and external divisions, GPi & GPe), nucleus accumbens (NAC), subthalamic nucleus (STN), substantia nigra (SN) (substantia nigra pars compacta (SNc) & substantia nigra pars reticulata (SNr)) and red nucleus (RN). 

Cassandra M.J. Wannan¹, Shane Tonissen², Bruce Tonge³, Efstratios Skafidas ^1,2, Christos Pantelis^1,2, and Warda T. Syeda^1
1Melbourne Neuropsychiatry Centre, The University of Melbourne, Parkville, Australia, ²Department of Electrical and Electronic Engineering, The University of Melbourne, Parkville, Australia, ³Psychiatry Monash Health, Faculty of Medicine, Nursing and Health Sciences School of Clinical Sciences, Monash University, Melbourne, Australia

A multi-contrast brain imaging technique has been presented, that leverages information from T₁- and T₂-weighted images to register iso-intense paediatric MRI data. The proposed method is applied to T₁-weighted images of 6-months old infants, and the performance is evaluated in comparison to commonly employed single-contrast registration techniques.

Jordan Wang¹, Po-Wei Cheng¹, Ming-Jang Chiu², Tzi-Dar Chiueh³, and Jyh-Horng Chen^3
1Graduate institute of biomedical electronics and bioinformatics, National Taiwan University, Taipei, Taiwan, ²National Taiwan University Hospital, Taipei, Taiwan, ³Department of electrical engineering, National Taiwan University, Taipei, Taiwan

The Single-frequency Excitation Wideband MRI (SE-WMRI) technique was implemented on a Siemens PRISMA 3T human MRI system to demonstrate its multifaceted benefits for human brain imaging, including acceleration, SNR improvement, and spatial resolution enhancement. The accelerated Wideband brain results showed high consistency with standard scans, and the resolution enhanced Wideband images revealed more detailed brain structures details and clearer blood vessel due to reduced partial volume effect. Since SE-WMRI can enhance MR efficiency without any hardware upgrade, it’s highly cost-effective for medical institutions in pursuit of higher MR imaging quality.

Suzie Bash¹, Long Wang², Chris Airriess³, Sara Dupont², Greg Zaharchuk⁴, Enhao Gong², Tao Zhang², Ajit Shankaranarayanan², and Lawrence Tanenbaum^5
1Neuroradiology, RadNet, Woodland Hills, CA, United States, ²Subtle Medical, Menlo Park, CA, United States, ³Cortechs.ai, San Diego, CA, United States, ⁴Stanford University, Stanford, CA, United States, ⁵RadNet, New York, NY, United States

In this prospective, multireader, multicenter study, we explore the impact of deep learning (DL) enhancement of 60% accelerated 3D T1 weighted brain MR image acquisitions.  We found that the DL processed images demonstrated high volumetric quantification accuracy, matched clinical disease status predictability, and provided what readers perceived as superior image quality when compared with the longer standard-of-care exams, suggesting good generalizability, accuracy, and potential utility of DL enhancement in routine clinical settings. The results of this trial support the use of DL enhancement to shorten clinical MR brain examinations, even when additional quantitative tools such as volumetric analysis are applied.   

Hajime Yokota¹, Takayuki Sakai², Masami Yoneyama³, Yansong Zhao⁴, and Takashi Uno^1
1Department of Diagnostic Radiology and Radiation Oncology, Chiba University, Chiba, Japan, ²Department of Radiology, Eastern Chiba Medical Center, Togane, Japan, ³Philips Japan, Tokyo, Japan, ⁴Philips Healthcare, Cleveland, OH, United States

Multi-Interleaved X-prepared tse with inTUitive Relaxometry (MIXTURE) is a 3D TSE sequence that can produce high-resolution morphology images and T2 map simultaneously. The purpose of this study was to evaluate MIXTURE FLAIR for T2 measurement of the cranial nerves. MIXTURE FLAIR produced reliable T2 values in the fantom study. The variance of T2 value measured by MIXTURE FLAIR was relatively small, and the T2 value was significantly smaller than multi-echo TSE and MIXTURE T2 in the volunteer study. MIXTURE FLAIR enabled the T2 value measurement of the cranial nerves, which was not previously feasible.

Makoto Obara¹, Yoshitomo Kikuchi², Akio Hiwatashi², Alexander Fischer ³, Yuta Akamine¹, Tetsuo Ogino¹, Masami Yoneyama¹, Ronee Asad¹, Yu Ueda¹, Jihun Kwon¹, and Marc Van Cauteren^4
1Philips Japan, Tokyo, Japan, ²Departments of Clinical Radiology, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, ³Philips GmbH Innovative Technologies, Aachen, Germany, ⁴Philips Healthcare, Tokyo, Japan

Volume isotropic simultaneous interleaved bright- and black-blood examination (VISIBLE) is an established approach for brain metastasis screening. We investigated the potential of VISIBLE in deep learning model development. The results suggest VISIBLE’s usefulness for automatic metastasis detection.

Rajiv G Menon¹, Marcelo V.W. Zibetti¹, and Ravinder R. Regatte^1
1New York University Langone Health, New York, NY, United States

3D-T_1ρ has many useful biomedical applications but requires long data acquisition times. The goal of the study was to apply a fast data-driven optimization approach, bias- accelerated subset selection (BASS), to generate optimal sampling patterns (SPs) for compressed sensing (CS) reconstruction for brain 3D-T1ρ MRI. Five healthy volunteers were recruited and fully sampled (FS) Cartesian, 3D-T1ρ MRI was obtained. The performance of Poisson disc (PD) and optimized SP were compared using normalized root mean square error (NRMSE). The data-driven optimized SP provides upto 2 times (NRMSE=0.09 optimized SP vs 0.18 PD-SP) improvement in images at the highest AFs tested.

Warda T. Syeda¹, Bjørn H. Ebdrup^1,2,3, Cassandra M.J. Wannan¹, Micah Cearns¹, Rigas Soldatos¹, Antonia Merritt¹, Mahesh Jayaram ¹, Andrew Zalesky ¹, Jayachandra M. Raghava^2,4, Birgitte Fagerlund ², Egill Rostrup ², Birte Glenthøj ^2,3, Leigh A. Johnston^5,6, Chad Bousman ^1,7, Ian Everall⁸, Efstratios Skafidas ^1,9, and Christos Pantelis^1,9
1Melbourne Neuropsychiatry Centre, The University of Melbourne, Parkville, Australia, ²Center for Neuropsychiatric Schizophrenia Research and Center for Clinical Intervention and Neuropsychiatric Schizophrenia Research, Mental Health Centre Glostrup, Copenhagen University Hospital, Glostrup, Denmark, ³Faculty of Health and Medical Sciences, Department of Clinical Medicine, University of Copenhagen, Denmark, Denmark, ⁴Functional Imaging Unit, Department of Clinical Physiology, Nuclear Medicine and PET, Rigshospitalet, Glostrup, Denmark, ⁵Department of Biomedical Engineering, The University of Melbourne, Melbourne, Australia, ⁶Melbourne Brain Centre Imaging Unit, The University of Melbourne, Parkville, Australia, ⁷Departments of Medical Genetics, Psychiatry, and Physiology & Pharmacology, University of Calgary, Calgary, AB, Canada, ⁸Institute of Psychiatry, Psychology & Neuroscience, King's College London, London, United Kingdom, ⁹Electrical and Electronic Engineering Department, The University of Melbourne, Parkville, Australia

Partial least squares (PLS) methods enable identification of multimodal patterns of latent associations between neuroimaging, cognitive, functional and clinical measures. Here, we propose a multiblock PLS correlation (MB-PLS-C) technique to enable covariate representation in the latent space and an interpretation framework to assess results from PLS-C analyses in clinical contexts. We investigate latent structure-cognition patterns in a multivariate dataset of individuals with treatment-resistant schizophrenia and healthy controls using the proposed MB-PLS-C method, and compare with standard PLS-C with and without covariate adjustment through residualization. 

Tracy Ssali¹, Lucas Narciso^1,2, Justin Hicks^1,2, Matthais Günther³, Frank Prato^1,2, Udunna Anazodo^1,2, Elizabeth Finger⁴, and Keith St Lawrence^1,2
1Lawson Health Research Institute, London, ON, Canada, ²Department of Medical Biophysics, Western University, London, ON, Canada, ³Fraunhofer Institute for Medical Image Computing MEVIS, Bremen, Germany, ⁴Department of Clinical Neurological Sciences, Western University, London, ON, Canada

The ability of arterial spin labeling (ASL) to detect perfusion abnormalities in clinical populations can be limited by poor signal to noise and transit-time artefacts. This is evident in studies involving frontotemporal dementia patients, where reports on the diagnostic value of ASL have been inconsistent. This study presents a head-to-head comparison of regional hypoperfusion detected by ASL and PET with radiolabeled water (¹⁵O-water), the gold standard for measuring CBF. T-maps depicting hypoperfusion were generated using absolute and relative CBF. There was good agreement between regional hypoperfusion identified by ASL and ¹⁵O-water, particularly for relative CBF maps which reduced inter-subject variability. 

Rachael C Stickland¹, Kristina M Zvolanek^1,2, Stefano Moia^3,4, Apoorva Ayyagari^1,2, César Caballero-Gaudes³, and Molly G Bright^1,2
1Physical Therapy and Human Movement Sciences, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States, ²Biomedical Engineering, McCormick School of Engineering, Northwestern University, Evanston, IL, United States, ³Basque Center on Cognition, Brain and Language, Donostia, Spain, ⁴University of the Basque Country EHU/UPV, Donostia, Spain

Blood Oxygen Level Dependent (BOLD) signals can be modulated by the baseline vascular and metabolic state. Understanding how baseline vascular physiology relates to dynamic neurovascular processes will lead to more accurate interpretations of BOLD Cerebrovascular Reactivity (CVR) measurements. We investigated the relationship between baseline Cerebral Blood Flow (bCBF) and BOLD-CVR, generally reporting positive correlations. Optimizing for vascular delays, and modelling with simple breathing task data, can improve CVR correlations with bCBF. Future work should investigate individual differences and include larger samples.

Kilian Stumpf¹, Hanna Frantz¹, Patrick Metze¹, Thomas Hüfken¹, Tobias Speidel¹, and Volker Rasche^1
1Department of Internal Medicine II, Ulm University Medical Center, Ulm, Germany

MR acquisitions of the eyes are often complicated by eye motions leading to the occurrence of motion artifacts in the images. Using a tiny golden angle profile ordering scheme and a sliding window reconstruction the position of the optic nerve is identified and tracked via peak analysis of signal intensities along a 1-D-line containing the optic nerve. An image-based self-gating signal is derived from the calculated nerve locations. This approach not only leads to distinct reduction of motion artifacts but also enables the visualization of various motion phases occurring during the data acquisition without the use of external tracking devices.

Junghun Cho¹, Pascal Spincemaille¹, Thanh D Nguyen¹, Ajay Gupta¹, and Yi Wang^1,2
1Radiology, Weill Cornell Medicine, New York, NY, United States, ²Biomedical Engineering, Cornell University, Ithaca, NY, United States

A Combined Cluster analysis of time evolution and tissue type with Total Variation denoising (CCTV) was developed to suppress noise propagation in oxygen extraction fraction (OEF) maps based on the QSM+qBOLD (QQ) model of multi-echo gradient echo data without vascular challenge. Compared to cluster analysis of time evolution (CAT), the developed CCTV provided more accurate OEF in simulation and greater contrast to noise ratio between lesion and its healthy contralateral side in ischemic stroke patients. 

Melanie Bauer^1,2, Celine Berger^1,2, Claudia Lenz^1,2, Eva Scheurer^1,2, and Christoph Birkl^3
1Institute of Forensic Medicine, Biomedical Engineering, University of Basel, Basel, Switzerland, ²Institute of Forensic Medicine, Health Department Basel-Stadt, Basel, Switzerland, ³Department of Neuroradiology, Medical University of Innsbruck, Innsbruck, Austria

To generate precise diffusion images in post mortem MRI, higher b-values are necessary than for in vivo MRI due to decreased diffusion. In this study, the influence of different b-values on the fiber orientation dependency of R₂* was assessed in 12 post mortem and 4 in vivo cases. Our results show that R₂* values are higher but their orientation dependency is lower post mortem than in vivo. However, the chosen b-values did not affect the estimation of the white matter fiber angle and, hence, the orientation dependency of R₂* for both post mortem in situ and in vivo MRI.

Steen Moeller¹, Cheryl Olman², Luca vizioli¹, Logan Dowdle¹, Essa Yacoub¹, Mehmet Akcakaya^1,3, and Kamil Ugurbil^1
1University of Minnesota, MINNEAPOLIS, MN, United States, ²Psychology, University of Minnesota, MINNEAPOLIS, MN, United States, ³ELECTRICAL AND COMPUTER ENGINEERING, University of Minnesota, Minneapolis, MN, United States

Investigating the utility of using the recently proposed NORDIC denoising prior to GRAPPA based unaliasing, for establishing the feasibility of integration with deep learning image reconstruction techniques.

Ken Sakaie¹, Janel Fedler², Jon Yankey², Kunio Nakamura¹, Josef Debbins³, Mark J. Lowe¹, Paola Raska¹, and Robert J. Fox^1
1The Cleveland Clinic, Cleveland, OH, United States, ²University of Iowa, Iowa City, IA, United States, ³Barrow Neurological Institute, Phoenix, AZ, United States

There is an urgent need for imaging biomarkers to develop new therapies for progressive multiple sclerosis. Hardware upgrades can confound the outcomes of imaging in clinical trials of such therapies. We examine different analytic approaches to evaluating the impact of and correcting for the effects of hardware changes in a retrospective analysis of SPRINT-MS, a multi-center clinical trial. Brain parenchymal fraction (BPF), a measure of atrophy, and transverse diffusivity (TD), a measure of demyelination are examined.

Ping Wang¹, Nicholas Sisco¹, and Richard Dortch^1
1Barrow Neurological Institute, Phoenix, AZ, United States

There is a significant need for specific biomarkers of myelin damage and repair. Quantitative magnetization transfer (qMT) using selective inversion recovery (SIR) is able to quantify the macromolecular-to-free proton pool size ratio (PSR), which has been shown to relate closely with myelin content and disability. For clinical applications, SIR is often hampered by long scan times. As a result, we employed compressed sensing and parallel imaging to significantly reduce scan times, with little effect on the precision and accuracy of PSR estimates in white matter.

Maarten Naeyaert¹, Tim Vanderhasselt¹, Marcel Warntjes², and Hubert Raeymaekers^1
1Radiology, Universitair Ziekenhuis Brussel, Brussels, Belgium, ²SyntheticMR AB, Linköping, Sweden

To evaluate intra- and inter-scan repeatability of quantitative scans, T1-, T2- and PD maps were acquired simultaneously using the 2D multi-delay-multi-echo sequence and the 3D-QALAS sequence, for 3T scanners of three different manufacturers. All scans were acquired with and without acceleration factor of 2, using GRAPPA, SENSE or compressed sensing, depending on the vendor. On one scanner measurements were repeated 10 times. The estimated values using synthetic MRI were compared to the reference values of the ISMRM-NIST phantom. The results for T1- and T2- mapping show a linear curve close to the reference values, with good reproducibility.

Bradley Fitzgerald¹, Kausar Abbas², Thomas M. Talavage^1,3, and Joaquin Goni^2
1Electrical & Computer Engineering, Purdue University, West Lafayette, IN, United States, ²Industrial Engineering, Purdue University, West Lafayette, IN, United States, ³Biomedical Engineering, University of Cincinnati, Cincinnati, OH, United States

Analysis of scan-rescan subject identifiability of T1 anatomical brain MRI could provide insight into intersession differences. Here we examine a principal component analysis-based method of maximizing differential identifiability and its effect on T1 brain images with and without added noise. We demonstrate that differential identifiability can be maximized via dataset reconstruction with reduced principle components. This reconstruction results in increased similarity between repeated scans for a given subject as well as apparent reduced intersession noise in images. We conclude that further analysis of maximized differential identifiability could provide insight for future applications in reducing intersession MRI noise.

Alina Lopatina^1,2, Stefan Ropele³, Renat Sibgatulin¹, Jürgen R Reichenbach^1,2,4, and Daniel Güllmar^1
1Medical Physics Group / IDIR, Jena University Hospital, Jena, Germany, ²Michael-Stifel-Center for Data-Driven and Simulation Science, Jena, Germany, ³Department of Neurology, Medical University of Graz, Graz, Austria, ⁴Center of Medical Optics and Photonics Jena, Jena, Germany

We propose a method to transform susceptibility-weighted images of multiple sclerosis (MS) patients to images reflecting healthy volunteers based on generative adversarial networks (GANs). This method helps to identify MS by changing voxel information corresponding to the disease. The results showed that voxels around the central veins and ventricles are identified as MS-specific by the method. This finding may contribute to improvements in MS diagnosis and encourage future studies based on the presented findings.

Anders Tisell^1,2, Peter Lundberg^1,2, Marcel Jan Bertus Warntjes¹, and Frederik Testud^3
1CMIV, Linköping University, Linköping, Sweden, ²Medical Radiation Physics, Linköping University, Linköping, Sweden, ³Siemens Healthcare AB, Malmö, Sweden

We have implemented the 3D QALAS sequence for a 3 T MR-scanner and validated the accuracy and evaluated the effect of B1+ inhomogeneities on the resulting R1 and R2 estimations. When using a small flip angle of 4° the determined R1 and R2 maps showed a high accuracy; however, for 10° there was a significant bias and dependence on B1+ inhomogeneities.