Muhamed Barakovic^1,2,3, Riccardo Galbusera^1,2,3, Reza Rahmanzadeh^1,2,3, Matthias Weigel^1,2,3, Po-Jui Lu^1,2,3, Erik Bahn⁴, Simona Schiavi⁵, Alessandro Daducci⁵, Pascal Sati^6,7, Pietro Maggi^8,9, Ludwig Kappos^2,3, Jens Kuhle^2,3, Laura Gaetano¹⁰, Stefano Magon¹¹, and Cristina Granziera^1,2,3
1Translational Imaging in Neurology (ThINk) Basel, University Hospital Basel and University of Basel, Basel, Switzerland, ²Neurologic Clinic and Policlinic, University Hospital Basel and University of Basel, Basel, Switzerland, ³Research Center for Clinical Neuroimmunology and Neuroscience (RC2NB) Basel, University Hospital Basel and University of Basel, Basel, Switzerland, ⁴Institute of Neuropathology, University Medical Center, Göttingen, Germany, ⁵Department of Computer Science, University of Verona, Verona, Italy, ⁶Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, Bethesda, MD, United States, ⁷Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ⁸Department of Neurology, Lausanne University Hospital, Lausanne, Switzerland, ⁹Cliniques universitaires Saint Luc, Université catholique de Louvain, Louvain, Belgium, ¹⁰F. Hoffmann-La Roche Ltd., Basel, Switzerland, ¹¹Pharmaceutical Research and Early Development, Roche Innovation Center Basel F. Hoffmann-La Roche Ltd., Basel, Switzerland

Paramagnetic rim lesions (PRL) are characterized by heterogeneous damage in multiple sclerosis (MS) patients. In this study we acquired in vivo, postmortem and histopathology with the aim to further characterize the microstructural properties of these lesions. Furthermore, we disentangled the heterogeneity of PRL by combining a stick-ball-sphere diffusion model with the quantification of mean T1 relaxation times.

Hang Zhang¹, Jinwei Zhang¹, Pascal Spincemaille¹, Thanh D. Nguyen¹, and Yi Wang^1
1Cornell University, New York, NY, United States

We propose an anatomical convolutional module to couple anatomical information into deep neural network. We further develop a loss function based on the mass center of individual lesions called lesion-wise loss, which can regularize the network training, thereby improving the performance of lesion localization and segmentation. We validate our methods on a public dataset, ISBI-15 Multiple Sclerosis Lesion Segmentation Challenge [1], where the results showed that we achieved the best performance on all published methods.

Zahra Hosseinpour¹, Olayinka Oladosu², Mahshid Soleymani¹, G Bruce Pike², and Yunyan Zhang^2
1Biomedical Engineering program, Schulish School of Engineering, University of Calgary, Calgary, AB, Canada, ²Department of Radiology and Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada

Multiple sclerosis (MS) lesion characterization using MRI is valuable for monitoring MS and probing remyelination therapies. However, there is a lack of in-vivo MRI metrics in clinical studies. We assessed MS lesion types using statistical texture measures of advanced and conventional MRIs, accompanied by random forest classification and percentile statistics, to differentiate MS lesions based on myelination. The best texture parameters and percentiles that classified re- and de-myelinated lesions were derived from histology-verified MRI.  Applying the established parameters from postmortem to in-vivo MRI identified two types of lesions: highly demyelinated and potentially remyelinated. 

Stephen T. Vito¹, Kai H. Barck², Rohan S. Virgincar², Amy Easton¹, Robby M. Weimer², Tracy J. Yuen¹, and Luke Xie^2
1Neuroscience, Genentech, South San Francisco, CA, United States, ²Biomedical Imaging, Genentech, South San Francisco, CA, United States

Multiple sclerosis (MS) is the most common disease of the CNS that causes neurodegeneration and demyelination. MRI can track changes in MS lesions, but how imaging monitors inflammatory cell response or myelin content is not well characterized. In this study, we use a cuprizone toxin mouse model to determine how T₂ relaxation time, myelin water fraction (MWF), diffusion tensor imaging (DTI), and gRatio MRI relate to changes with microglia, astrocytes, oligodendrocytes, and myelin. Using longitudinally imaging, MRI was able to detect inflammation-mediated demyelination. T₂ reflected changes in inflammation (microglia and astrocytes) while MWF and gRatio suggested demyelination.

Maxence Wynen¹, Francesco La Rosa^2,3,4, Amina Sellimi⁵, Germán Barquero^2,3,4, Gaetano Perrotta⁶, Valentina Lolli⁷, Vincent Van Pesch⁵, Cristina Granziera^8,9, Tobias Kober¹⁰, Pascal Sati^11,12, Benoît Macq¹³, Daniel S. Reich¹¹, Martina Absinta^11,14, Meritxell Bach Cuadra^2,3,4, and Pietro Maggi^5,15
1Ecole Polytechnique de Louvain, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, ²Signal Processing Laboratory (LTS5), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ³Medical Image Analysis Laboratory, Center for Biomedical Imaging (CIBM), University of Lausanne, Lausanne, Switzerland, ⁴Radiology Department, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ⁵Department of Neurology, Cliniques Universitaires Saint-Luc, Université Catholique de Louvain, Brussels, Belgium, ⁶Department of Neurology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, ⁷Department of Radiology, Erasme University Hospital, Université Libre de Bruxelles, Brussels, Belgium, ⁸Neurologic Clinic and Policlinic, Departments of Medicine, Clinical Research and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, ⁹Translational Imaging in Neurology (ThINk) Basel, Department of Medicine and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, ¹⁰Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, ¹¹Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, United States, ¹²Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ¹³ICTEAM Institute, Université catholique de Louvain, Louvain-la-Neuve, Belgium, ¹⁴Department of Neurology, Johns Hopkins University, Baltimore, MD, United States, ¹⁵Department of Neurology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland

The automatic assessment of paramagnetic rim lesions in multiple sclerosis is important, and a deep learning-based algorithm called RimNet has recently been proposed. This work evaluates the generalizability of RimNet and its longitudinal performance on MRI data acquired at different clinical centers. We found that RimNet’s performance was nearly as good on totally unseen data as in the original paper (receiver-operating-characteristic area-under-the-curve (AUC) 0.88 vs. 0.94, precision-recall AUC 0.69 vs. 0.70), and it made consistent predictions on longitudinal data (binary consistency 82%, probability consistency 93%).

Quan Chen¹, Huajun She¹, and Yiping P. Du^1
1School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China

Acceleration of myelin water fraction (MWF) mapping using the Feature domain nonlocal Low-Rank Tensor based (FnLRT) algorithm is investigated in this study. The global temporal information of the whole images is used to project the T2* weighted images (T2*WIs) into the feature domain. The nonlocal and local spatial redundancies in the feature domain are further exploited. The tensor-based decomposition is used to explore the multi-dimensional redundancies. The human brain experiments demonstrate the outperformance of the FnLRT algorithm over the state-of-the-art reconstructions at R=6. The FnLRT algorithm provides the potential to obtain the whole brain MWF mapping in 1 minute. 

Poljanka Johnson¹, Irene M Vavasour², Shawna Abel¹, Lisa E Lee³, Stephen Ristow¹, Cornelia Laule⁴, Roger Tam², David K.B. Li², Nathalie Ackermans¹, Alice Schabas¹, Jillian Chan¹, Helen Cross¹, Ana-Luiza Sayao¹, Virginia Devonshire¹, Robert Carruthers¹, Anthony Traboulsee³, and Shannon Kolind^5
1Medicine, University of British Columbia, Vancouver, BC, Canada, ²Radiology, University of British Columbia, Vancouver, BC, Canada, ³University of British Columbia, Vancouver, BC, Canada, ⁴Radiology, Physics and Astronomy, Pathology, University of British Columbia, Vancouver, BC, Canada, ⁵Medicine, Physics & Astronomy, University of British Columbia, Vancouver, BC, Canada

Myelin water imaging is a quantitative MRI technique that has been used to investigate myelin content in multiple sclerosis (MS). Three metrics derived from myelin water imaging were compared in the normal appearing white matter of different MS subtypes and correlations with disability were investigated. The myelin heterogeneity index was best able to distinguish between MS subtypes and demonstrated the strongest correlation with disability.

Lucas Soustelle^1,2, Andreea Hertanu^1,2, Arnaud Le Troter^1,2, Soraya Gherib^1,2, Samira Mchinda^1,2, Patrick Viout^1,2, Lauriane Pini^1,2, Claire Costes^1,2, Sylviane Confort-Gouny^1,2, Adil Maarouf^1,2,3, Bertrand Audoin^1,2,3, Audrey Rico^1,2,3, Clémence Boutière^1,2,3, Maxime Guye^1,2, Jean-Philippe Ranjeva^1,2, Gopal Varma⁴, David C. Alsop⁴, Jean Pelletier^1,2,3, Olivier M. Girard^1,2, and Guillaume Duhamel^1,2
1Aix Marseille Univ, CNRS, CRMBM, Marseille, France, ²APHM, Hôpital Universitaire Timone, CEMEREM, Marseille, France, ³APHM, Hôpital Universitaire Timone, Service de neurologie, Marseille, France, ⁴Division of MR Research, Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States

Active lesions in Multiple Sclerosis (MS) present successive phases, from formation through their chronic stage. Characterizing these phases may help us better comprehend the disease evolution. In this work, multiparametric MRI was performed in a 12-month longitudinal study of MS patients presenting new active lesions. An exponential recovery model was proposed to characterize the evolution of MR metrics in these lesions, allowing derivation of the recovery rates of inhomogeneous MTR, conventional MTR, DTI and T₁-weighted images. Results show that recovery capacities are patient-dependent and that metrics differ in performance, presumably due to their respective sensitivity to the underlying MS mechanisms.

Kyle Jeong¹, Lubdha Shah², You-Jung Lee¹, Bijaya Thapa¹, Nabraj Sapkota¹, Erica Bisson³, Noel Carlson⁴, Eun-Kee Jeong², and John Rose^5
1Utah Center for Advanced Imaging Research, University of Utah, Salt Lake City, UT, United States, ²Radiology and Imaging Sciences, University of Utah, Salt Lake City, UT, United States, ³Neurosurgery, University of Utah, Salt Lake City, UT, United States, ⁴GRECC, Veterans Affairs, Salt Lake City, UT, United States, ⁵Neuroimmunology and Neurovirology Division, University of Utah, Salt Lake City, UT, United States

Ultrahigh-b radial DWI (UHb-rDWI) is a technique that can quantitatively evaluate MS lesions with respect to degrees of demyelination and axonal damage and/or loss. Currently, there is a lack of sensitive diagnostic technique to quantitatively and non-invasively grade cervical spinal cord (CSC) injury and repair. Our UHb-rDWI can potentially be a powerful tool to quantitatively observe disease progression and monitor treatment responses in MS CSC. Therefore, the main objective of this study was to investigate UHb-rDWI signal in white matter tracts of the CSC and compare quantitative values between healthy control WM with both MS NAWM and WM lesions.

Ya-Jun Ma¹, Hyungseok Jang¹, Zhao Wei¹, Mei Wu¹, Saeed Jerban¹, Eric Y Chang^1,2, Jody Corey-Bloom¹, Graeme M Bydder¹, and Jiang Du^1
1UC San Diego, San Diego, CA, United States, ²VA Health system, San Diego, CA, United States

We combine a Short TR Adiabatic Inversion Recovery preparation and dual-echo Ultrashort TE data collection with complex Echo Subtraction (ES) (STAIR-dUTE-ES) for both morphological and quantitative imaging of ultrashort T2 components in the whole brain. The 3D STAIR-dUTE-ES technique provides robust water suppression within the whole brain and allows accurate ultrashort T2 proton quantification. UltraShort T2 Proton Fraction (USPF) reduction in multiple sclerosis (MS) lesions suggests that the STAIR-dUTE-ES technique has potential for evaluation of demyelination and remyelination in the diagnosis and treatment of patients with MS.

Irene Margaret Vavasour¹, Poljanka Johnson², Shawna Abel³, Stephen Ristow³, Jared Splinter³, Cornelia Laule^1,4,5,6, Roger Tam¹, David KB Li¹, Nathalie Ackermans³, Alice J Schabas³, Jillian Chan³, Helen Cross³, Ana-Luiza Sayao³, Virginia Devonshire³, Robert Carruthers³, Anthony Traboulsee³, and Shannon H Kolind^1,3,4,6
1Radiology, University of British Columbia, Vancouver, BC, Canada, ²Neuroscience, University of British Columbia, Vancouver, BC, Canada, ³Medicine, University of British Columbia, Vancouver, BC, Canada, ⁴Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, ⁵Pathology and Laboratory Medicine, University of British Columbia, Vancouver, BC, Canada, ⁶International Collaboration on Repair Discoveries (ICORD), University of British Columbia, Vancouver, BC, Canada

Therapies that target remyelination are under development and a non-invasive specific and sensitive imaging biomarker to evaluate their efficacy within the timescale of a clinical trial is vital. Using myelin water imaging, we followed a group of relapsing-remitting and progressive multiple sclerosis (MS) participants and healthy controls over approximately 2 years to compare their rate of change in myelin water fraction. A significant decrease in mean myelin water fraction was found over 2 years in the normal-appearing white matter of participants with MS, with a larger decrease in relapsing-remitting MS than progressive MS.

Jeremy Kim^1,2, Thanh Nguyen², Jinwei Zhang², and Yi Wang^2
1Stanford University, New York, NY, United States, ²Weill Cornell, New York, NY, United States

We developed UNET neural network for rapid, accurate and reproducible extraction of myelin water fraction map from FAST-T2 multi-echo T2 decay data. Testing results on 109 MS brains showed that UNET shortens post-processing time to less than a second and outperforms existing multi-layer perceptron algorithm.

Sara Collorone¹, Marco Battiston¹, Ferran Prados ^1,2,3, Alberto Calvi¹, Baris Kanber⁴, Francesco Grussu^1,5, Marios Yiannakas¹, Carmen Tur^1,6, Rebecca Samson¹, Olga Ciccarelli^1,7, and Claudia A.M. Gandini Wheeler-Kingshott^1,8,9
1NMR Research Unit, Queen Square MS Centre, Department of Neuroinflammation, UCL Queen Square Institute of Neurology, Faculty of Brain Sciences, University College London (UCL), London, United Kingdom, ²Centre for Medical Image Computing (CMIC), Department of Medical Physics and Biomedical Engineering, University College London (UCL), London, United Kingdom, ³Universitat Oberta de Catalunya, Barcelona, Spain, ⁴Centre for Medical Image Computing (CMIC), Department of Medical Physics and Biomedical Engineering, University College London(UCL), London, United Kingdom, ⁵Radiomics Group, Vall d’Hebron Institute of Oncology, Vall d’Hebron Barcelona Hospital Campus, Barcelona, Spain, ⁶Multiple Sclerosis Centre of Catalonia (Cemcat), Vall d’Hebron Institute of Research, Vall d’Hebron Barcelona Hospital Campus, Barcelona, Spain, ⁷University College London Hospitals, Biomedical Research Centre, National Institute for Health Research, London, United Kingdom, ⁸Department of Brain & Behavioural Sciences, University of Pavia, Pavia, Italy, ⁹Brain Connectivity Center Research Department, IRCCS Mondino Foundation, Pavia, Italy

We present preliminary results on ten patients with progressive multiple sclerosis(PMS) who underwent multi-modal MRI including quantitative magnetisation transfer (qMT), relaxometry and neurite orientation dispersion and density imaging (NODDI). We obtained metrics sensitive to iron deposition (T2*), myelin (restricted proton fraction (F)), and neuro-axonal density (neurite density index (NDI)) in lesions and brain tissues. Increased lesion T2* was related to decreased NDI in lesions and cortical grey matter. The lesion high T2* and low NDI were related to worse cognitive performance. In PMS, iron deposition from innate immune activity may lead to axonal loss and represent a future therapeutic target.

Aly Khalifa¹ and Michael D Noseworthy^2,3
1Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada, ²Electrical and Computer Engineering, McMaster University, Hamilton, ON, Canada, ³School of Biomedical Engineering, McMaster University, Hamilton, ON, Canada

A data analysis pipeline was developed to correlate multiple sclerosis (MS) disease severity with regional susceptibility weighted imaging (SWI) intensity scores. It was assumed that reduced brain SWI signal correlated with elevated iron accumulation. Susceptibility weighted imaging (SWI) was used to map regional variations in brain iron content within each MS subtype. Number of brain areas with abnormal brain iron (relative to healthy age/sex matched controls), and the size of abnormal SWI image clusters, correlated with disease severity, based on subtype. 

Hang Zhang¹, Jinwei Zhang¹, Junghun Cho¹, Susan A. Gauthier¹, Pascal Spincemaille¹, Thanh D. Nguyen¹, and Yi Wang^1
1Cornell University, New York, NY, United States

Lesion count, which encodes the lesion historical information, is an important biomarker for diagnosis and treatment of multiple sclerosis. Confluent lesions pose a great challenge to traditional automated methods, as these lesions are connected spatially, which requires expert experience to separate them. In this abstract, we propose a Hough voting method based on deep neural networks to resolve the issue. Experimental results on an in-house dataset demonstrates the superiority of our approach.

Liangdong Zhou¹, Yi Li¹, Xiuyuan Hugh Wang¹, Elizabeth Sweeney¹, Hang Zhang^1,2, Emily B Tanzi¹, Jennette Prince², Victor Antonio Su-Ortiz², Susan A Gauthier¹, and Thanh D Nguyen^1
1Radiology, Weill Cornell Medicine, New York, NY, United States, ²Cornell University, Ithaca, NY, United States

Multiple Sclerosis (MS) includes both inflammatory demyelination and axonal degeneration, which may affect the tissue water fractions of brain parenchyma in both myelin and the CSF water compartments. Quantification of brain parenchyma CSF fraction (CSFF) may help us better understanding of this complicated neurodegenerative disease. Aging is the most common cause of neurodegenerations. Here we tested CSFF in a 111 non-active MS cohort (age:45-79 years) by using the multi-echo T2 relaxometry-based multiple water-compartment method. Multivariate multiple regression shows that in cortical gray matter CSFF increases with age by controlling other factors including gender, disease duration, and MS lesion burden.

Stefania Oliviero¹ and Cosimo Del Gratta^1
1Department of Neuroscience, Imaging, and Clinical Sciences, University of Chieti Pescara G. D'Annunzio, Chieti, Italy

There is a general lack of agreed-upon guidelines regarding the DWI sequences to use, particularly to optimize the microstructural characterization of lesions. We evaluated the effects of demyelination and axonal loss on DKI and NODDI metrics and the impact of the sequence on the sensitivity to damage by performing a Monte-Carlo diffusion simulation inside a novel model of damaged WM. The sequence strongly affected the means and sensitivities of the metrics. Consequently, comparing DKI and NODDI analysis employing different sequences could lead to erroneous conclusions regarding the damage assessment: further investigations are needed for a community consensus on acquisition details.

Emily Louise Baadsvik¹, Markus Weiger¹, Romain Froidevaux¹, Wolfgang Faigle², Benjamin Victor Ineichen², and Klaas Paul Pruessmann^1
1ETH Zurich and University of Zurich, Zurich, Switzerland, ²University Hospital Zurich and University of Zurich, Zurich, Switzerland

Ultrashort-T₂ signal from white matter can be captured for imaging using dedicated short-T₂ techniques. In this work, such methods were applied to D₂O-exchanged human brain tissue from four donors diagnosed with multiple sclerosis to obtain 0.39 mm isotropic resolution MR images of the ultrashort-T₂ tissue components and a lower-resolution multi-TE image series which, via a model fitting procedure, was used to produce maps of the relative content of myelin lipid-protein bilayer. As proof-of-principle, the contrast in these images was compared with corresponding myelin-stained cryosections and sample photographs, yielding promising correlation for white matter, grey matter and multiple sclerosis lesions.

Yuki Kanazawa¹, Masafumi Harada¹, Yo Taniguchi², Syun Kitano³, Nagomi Fukuda³, Yuki Matsumoto¹, Hiroaki Hayashi⁴, Kosuke Ito², Yoshitaka Bito², and Akihiro Haga^1
1Graduate School of Biomedical Sciences, Tokushima University, Tokushima, Japan, ²Healthcare Business Unit, Hitachi, Ltd., Tokyo, Japan, ³Faculty of Medicine, Tokushima University, Tokushima, Japan, ⁴Faculty of Health Sciences, Institute of Medical, Pharmaceutical and Health Sciences, Kanazawa University, Kanazawa, Japan

We evaluated myelin imaging R₁·R₂^* derived from QPM-relaxation parameters compared with the MWF derived from the same QPM gradient-echo dataset. Linear regression analysis indicated a correlation between R₁·R₂^* products values and MWF values derived from same QPM dataset (R = 0.53, P < 0.0001).The R₁·R₂^* products map derived from QPM offers as follow advantage; quantitative analysis; artifact free; and post-processing without fitting algorithm after QPM estimation. Myelin map derived from QPM can be applied for the quantitative assessment of white matter structure as substitute for MWF.

Tomoko Maekawa^1,2, Akifumi Hagiwara^1,3, Masaaki Hori^1,4, Christina Andica¹, Shohei Fujita^1,5, Toshiaki Akashi¹, Koji Kamagata¹, Akihiko Wada¹, and Shigeki Aoki^1
1Department of Radiology, Juntendo University School of Medicine, Tokyo, Japan, ²Division of Regenerative Medicine, The Jikei University School of Medicine, Tokyo, Japan, ³Department of Radiological Science, David Geffen School of Medicine, University of California Los Angeles, Los Angeles, CA, United States, ⁴Department of Diagnostic Radiology, Toho University Omori Medical Center, Tokyo, Japan, ⁵Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan

We investigated the association between estimated time from the onset of multiple sclerosis (MS) plaques and myelin- and axon-related quantitative MRI measurements using synthetic MRI (SyMRI) and neurite orientation dispersion and density imaging (NODDI). We retrospectively analyzed 32 MS patients with 73 newly appearing plaques. MS plaques with longer estimated time from the onset had significantly lower myelin volume fraction and higher g-ratio. It is suggested that myelin in plaques is under ongoing damage more than axons. SyMRI and NODDI may be useful for the quantitative assessment of temporal change in MS plaques.