Llucia Coll¹, Pere Carbonell-Mirabent¹, Álvaro Cobo-Calvo¹, Georgina Arrambide¹, Ángela Vidal-Jordana¹, Manuel Comabella¹, Joaquín Castilló¹, Breogán Rodríguez-Acevedo¹, Ana Zabalza¹, Ingrid Galán¹, Luciana Midaglia¹, Carlos Nos¹, Annalaura Salerno², Cristina Auger², Manel Alberich², Jordi Río¹, Jaume Sastre-Garriga¹, Arnau Oliver³, Xavier Montalban¹, Àlex Rovira², Mar Tintoré¹, Deborah Pareto², Xavier Lladó³, and Carmen Tur^1
1Multiple Sclerosis Centre of Catalonia (Cemcat), Department of Neurology/Neuroimmunology, Hospital Universitari Vall d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain, ²Section of Neuroradiology, Department of Radiology (IDI), Vall d'Hebron University Hospital, Spain. Universitat Autònoma de Barcelona, Barcelona, Spain, ³Research institute of Computer Vision and Robotics, University of Girona, Girona, Spain

The ability of existing MRI biomarkers to predict MS patients’ prognosis is limited and inaccurate to be used at the individual level. We aimed to assess the ability of Convolutional Neural Networks (CNNs) to classify relapse-onset MS patients into non-disabled and markedly disabled using only MRI images. T1w and T2w-FLAIR images of 538 MS patients were used to train and test two CNN approaches which were compared also with (conventional) logistic regression models. The results showed that the CNN models performed better, having the intrinsic potential to improve after the inclusion of regional priors and other valuable clinical data.

Christian Federau¹, Guangming Zhu², Nicolin Hainc³, Silvio Paganucci¹, Lukas Kipp², and Max Wintermark^2
1AI Medical, Zollikon, Switzerland, ²Stanford University, Stanford, CA, United States, ³University Zürich, Zürich, Switzerland

Yearly multiple sclerosis MRI follow-ups require the tedious and time-consuming manual comparing and counting of multiple demyelinating lesions, which can reach hundreds in some cases. We evaluated a semi-automatic software for the efficient assessment of such images, and found that a significant increase in the number of reported new multiple sclerosis lesions can be achieved in two-and-a-half minutes reading on average per case. In contrast, current standard reports missed 80% of the new lesions. In addition, The Jazz software permits the efficient reporting of slowly evolving lesions, an entity growing in clinical relevance and typically overlooked in current reporting. 

Sirio Cocozza¹, Renato Cuocolo¹, Giuseppe Pontillo¹, Lorenzo Ugga¹, Maria Petracca¹, Teresa Costabile¹, Roberta Lanzillo¹, Vincenzo Brescia Morra¹, Mario Quarantelli², and Arturo Brunetti^1
1University of Naples Federico II, Naples, Italy, ²National Research Council, Naples, Italy

The identification of early biomarkers to predict the disability accumulation is crucial in Multiple Sclerosis (MS). We performed a combined radiomics and Machine Learning (ML) study to predict long-term clinical changes in MS. Radiomics data were extracted from data of 177 patients with a 10-years clinical follow-up available. The model based on the recursive elimination of the features combined with the Extra Trees classifier was able to obtain a maximum precision for each endpoint of 0.71 and 0.69 for cognitive and motor disability, respectively.This combined radiomics-ML approach seems to be a feasible tool for long-term clinical prediction in MS.

Gawon Lee¹, Ji Wan Son¹, Ken SaKaie², Kunio Nakamura³, Yufan Zheng³, Daniel Ontaneda⁴, Bruce Trapp⁵, Mark Lowe², Dong Hye Ye⁶, and Se-hong Oh^7
1Division of Biomedical engineering, Hankuk University of Foreign Studies, Yongin-si, Gyeonggi-do, Korea, Republic of, ²Imaging institute, Cleveland Clinic, Cleveland, OH, United States, ³Biomedical Engineering, Lerner Research Institue, Cleveland Clinic, Cleveland, OH, United States, ⁴Mellen Center for Multiple Sclerosis, Cleveland Clinic, Cleveland, OH, United States, ⁵Department of Neurosciences, Cleveland Clinic, Cleveland, OH, United States, ⁶Department of Electrical and Computer Engineering, Marquette University, Milwaukee, WI, United States, ⁷Biomedical engineering, Hankuk University of Foreign Studies, Yongin-si, Gyeonggi-do, Korea, Republic of

When a clinical MR scan is acquired, there might be missing tissue contrasts due to the corruption by patient’s motion during long scan time. In this study, we propose a method to synthesize the missing T2-weighted or FLAIR contrasts from a T1-weighted image using physics-constrained neural network. We incorporate the Bloch equations that generate MR contrast images from tissue parameter maps based on MR physical models into a synthesis neural network and show the improved performance compared with the existing U-Net directly synthesizing from a T1-weighted image.

Simona Schiavi¹, Gian Franco Piredda^2,3,4, Caterina Lapucci⁵, Francesco Tazza¹, Domenico Zacà⁶, Luca Roccatagliata^7,8, Tom Hilbert^2,3,4, Tobias Kober^2,3,4, Matilde Inglese^1,9, and Mauro Costagli^1
1Department of Neuroscience, Rehabilitation, Ophthalmology, Genetics, Maternal and Child Health (DINOGMI), University of Genoa, Genoa, Italy, ²Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, ³Department of Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ⁴LTS5, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ⁵HNSR, IRRCS Ospedale Policlinico San Martino, Genoa, Italy, ⁶Siemens Healthcare s.r.l, Milan, Italy, Milan, Italy, ⁷Department of Neuroradiology, IRCCS Ospedale Policlinico San Martino, Genoa, Italy, ⁸Dipartimento di Scienze della Salute (DISSAL), University of Genoa, Genoa, Italy, ⁹IRCCS Ospedale Policlinico San Martino, Genoa, Italy

Using a comprehensive multimodal quantitative MRI protocol, we study the relationships between different microstructural metrics inside lesional tissue and investigate the heterogeneous pathological processes underlying alterations visible as hyperintense plaques in FLAIR images. Our preliminary results suggest that, although all the metrics can detect differences between lesions and normal appearing white matter, not all are directly associated with clinical status. Moreover, many metrics are intercorrelated and should not be considered as independent information when analyzing clinical outcomes. Understanding the clinical value of these parameters can advance the understanding of complex microstructural processes in neurological diseases and informs MRI protocol designs.

Chang-Woo Ryu¹, Xiao-Yi Guo², HyeokJung Kwon², Soonchan Park¹, Hak Young Rhee³, A-Rang Cho⁴, and Geon-Ho Jahng^1
1Radiology, Kyung Hee University Hospital at Gangdong, Seoul, Korea, Republic of, ²Medicine, Kyung Hee University, Seoul, Korea, Republic of, ³Neurology, Kyung Hee University Hospital at Gangdong, Seoul, Korea, Republic of, ⁴Psychiatry, Kyung Hee University Hospital at Gangdong, Seoul, Korea, Republic of

To assess the cerebral microvascular alteration in the demented patients relative to a non-demented population using in vivo microvascular index maps by using Gd-based contrast agent at a 3T MRI system, we included 11 non-demented participants and 11 AD patients. Compared with the non-demented group, BVf and MvWI were significantly increased in the demented group. Both mVD and VSI were only significantly decreased in the demented group at the white matter hyperintensity (WMHI) area. BVf and MvWI were significantly positively correlated with age and VSI was significantly positively correlated with MMSE.

Cristian Montalba^1,2,3, Pamela Franco^1,3,4, Tomás Labbé⁵, Marcelo Andia^1,2, Miguel Guevara⁶, Jean-François Mangin⁷, Juan Pablo Cruz², Ethel Ciampi^8,9, Claudia Cárcamo^5,8, Pamela Guevara⁶, and Sergio Uribe^1,2,3
1Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile, ²Radiology Department, School of Medicine, Pontificia Universidad Católica de Chile, Santiago, Chile, ³Millennium Nucleus for Cardiovascular Magnetic Resonance, Santiago, Chile, ⁴Electrical Engineering Department, School of Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁵Interdisciplinary Center of Neurosciences, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁶Faculty of Engineering, Universidad de Concepción, Concepción, Chile, ⁷UNATI, Neurospin, CEA, Université Paris-Saclay, Gif-sur-Yvette, France, ⁸Neurology Department, School of Medicine, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁹Neurology Service, Hospital Dr. Sótero del Río, Santiago, Chile

Radiological biomarkers of cognitive impairment in Multiple Sclerosis (MS) is still scarce. This study aimed to identify subcortical white matter biomarkers of cognitive impairment related to verbal episodic memory in MS patients and healthy controls, by using a random forest approach.

Vanessa Wiggermann¹, Mie Arnau Martinez¹, Jasmin Merhout¹, Jeppe Romme Christensen², Morten Blinkenberg², Finn Sellebjerg^2,3, Henrik Lundell^1,4, and Hartwig Roman Siebner^1,5
1Danish Research Centre for Magnetic Resonance, Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital - Amager and Hvidovre, Hvidovre, Denmark, ²Neurology, Danish Multiple Sclerosis Center, Rigshospitalet, Glostrup, Denmark, ³Clinical Medicine, University of Copenhagen, Copenhagen, Denmark, ⁴Health Technology, Technical University of Denmark, Lyngby, Denmark, ⁵Neurology, Copenhagen University Hospital Bispebjerg, Copenhagen, Denmark

Black holes (BHs), T1 hypointense lesions, are known predictors of multiple sclerosis disease severity, however, their identification is strongly intensity dependent. As traditional spin-echo T1 scans are being replaced by 3D gradient-echo images, we need to understand how hypointense lesions on these images compare. We assessed whether an intensity threshold for MPRAGE images exists that would allow to match spin-echo black hole definitions. Simple thresholding was only able to reproduce black hole segmentations to an intermediate degree.

Maria Petracca^1,2, Alberto Azzari³, Antonella Mensi³, Nicole Graziano², Alessandro Daducci³, Manuele Bicego³, Matilde Inglese^2,4,5,6, and Simona Schiavi^3,4
1Department of Human Neurosciences, Sapienza University of Rome, Rome, Italy, ²Department of Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, ³Department of Computer Science, University of Verona, Verona, Italy, ⁴Department of Neuroscience, Rehabilitation, Ophthalmology, Genetics, Maternal and Child Health (DINOGMI), University of Genoa, Genoa, Italy, ⁵Department of Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, ⁶IRCCS Ospedale Policlinico San Martino, Genoa, Italy

Improving image-based classification accuracy in multiple sclerosis while characterizing biological relevant features remains an open challenge. To this aim we applied a robust feature selection (FS) procedure based on a leave-one-out cross-validation scheme on structural connectivity features derived from connectomes computed with convex optimization modelling for microstructure informed tractography. We computed classification accuracy for different classifiers (NN, KNN, SVM-LIN, SVM-RBF, RF) before and after the application of the FS procedure. The highest overall accuracy (91%) was obtained using the FS procedure on the whole connectome. The biological meaningfulness of the selected features is supported by their correlations with clinical scores.

Yujing Li¹, Jun Wang¹, Tingli Yang¹, Pengfei Zhang¹, Kai Ai², Xiaoli He¹, Haiying Zhao¹, Yuan Ding¹, Rui Wang¹, Min Li¹, Xinying Ren¹, Diaohan Xiong¹, and Jing Zhang^1
1Department of Magnetic Resonance, Lanzhou University Second Hospital, Lanzhou, China, ²Philips Healthcare, Xi’an, China

The objective of this study was to elucidate the no-motor component of fatigue mechanism in multiple sclerosis (MS) patients by studying the resting state functional connectivity (RS-FC) of default mode network (DMN). 12 with fatigue (F) and 10 without fatigue (nF) MS underwent resting sate magnetic resonance imaging (MRI). We selected 11 priori regions within the DMN as the regions of interests (ROIs) and ROI to ROI and ROI to global FC were calculated. Compared to nF-MS patients, F-MS patients showed abnormal increased functional connectivity in brain regions associated with cognitive function within the DMN.

Giuseppe Pontillo¹, Maria Petracca¹, Serena Monti², Mario Quarantelli², Roberta Lanzillo¹, Teresa Costabile³, Antonio Carotenuto¹, Fabio Tortora¹, Andrea Elefante¹, Vincenzo Brescia Morra¹, Arturo Brunetti¹, Giuseppe Palma², and Sirio Cocozza^1
1University of Naples "Federico II", Naples, Italy, ²Institute of Biostructure and Bioimaging, National Research Council, Naples, Italy, ³Università degli Studi della Campania "Luigi Vanvitelli", Naples, Italy

In this cross-sectional study, voxel-based morphometry and voxel-based quantification analyses of longitudinal relaxation rate (R1) and quantitative susceptibility (QSM) maps, reflecting myelin and iron, respectively, were conducted on 117 patients with MS (pwMS) and 53 healthy controls. Between-group differences and correlations with clinical variables (global disability, cognitive and motor performance) were assessed. We found widespread atrophy and R1 decrease in pwMS, with smaller clusters of reduced susceptibility in the thalami and cerebral WM. While subcortical atrophy and WM iron depletion contributed to global and motor disability, deep GM atrophy and R1 decrease in lesional areas correlated with cognitive performance.

Elizabeth York¹, Rozanna Meijboom¹, Michael J. Thrippleton¹, Agniete Kampaite¹, Maria Valdes Hernandez¹, Peter Connick², Siddharthan Chandran^1,2, David P.J. Hunt¹, and Adam D. Waldman^1
1Centre for Clinical Brain Sciences, University of Edinburgh, Edinburgh, United Kingdom, ²Anne Rowling Regenerative Neurology Clinic, University of Edinburgh, Edinburgh, United Kingdom

Myelin-sensitive magnetisation transfer saturation (MTsat) corrects magnetisation transfer ratio (MTR) signal variance, accounting for T1 recovery and B1 inhomogeneities, but its validity as a longitudinal imaging marker in multiple sclerosis (MS) is not yet clear. Here, we examine longitudinal change in cerebral MTsat and MTR in a cohort of people with newly diagnosed MS  (n=62). MTsat, but not MTR, decreased in normal-appearing white matter over one year and was associated with clinical disability progression. In white matter lesions, both MTsat and MTR increased longitudinally. We conclude that MTsat is more sensitive to subtle myelin damage over time than MTR.

Lisa Eunyoung Lee^1,2, Irene Vavasour³, Enedino Hernandez-Torres⁴, Anthony Traboulsee⁵, Roger Tam^3,5,6, Shannon Kolind^3,5,7, Tom Schweizer^2,8, and Jiwon Oh^1,2
1Medicine (Neurology), University of Toronto, Toronto, ON, Canada, ²Medicine, Keenan Research Centre for Biomedical Research, Toronto, ON, Canada, ³Radiology, University of British Columbia, Vancouver, BC, Canada, ⁴Copenhagen University Hospital, Copenhagen, Denmark, ⁵Medicine (Neurology), University of British Columbia, Vancouver, BC, Canada, ⁶Biomedical Engineering, University of British Columbia, Vancouver, BC, Canada, ⁷Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, ⁸Medicine (Neurosurgery), University of Toronto, Toronto, ON, Canada

Magnetization transfer saturation (MTsat) is suggested to be superior to conventional MT ratio (MTR) in quantifying myelin as it reduces bias from T1 effect and B1 inhomogeneity and is more sensitive to tissue damage. This study aimed to compare the ability of MTsat and MTR to detect cerebellar myelin differences across multiple sclerosis (MS) subtypes and to evaluate whether MTsat or MTR demonstrates stronger correlations with clinical disability. We found that MTsat demonstrated greater group differences and more strongly correlated with clinical disability in cerebellum than MTR suggesting that MTsat may be more sensitive to cerebellar myelin abnormalities than MTR.

Ronja C. Berg¹, Thomas Amthor², Irene Vavasour³, Mariya Doneva², and Christine Preibisch^1
1Department of Neuroradiology, School of Medicine, Technical University of Munich, Munich, Germany, ²Philips Research Europe, Hamburg, Germany, ³Department of Radiology, University of British Columbia, Vancouver, BC, Canada

Myelin water fraction (MWF) mapping provides information on myelin concentration, which can be affected by neurological diseases. Most commonly, a non-negative least squares (NNLS) algorithm is used to obtain MWF. A faster alternative is the Sparsity Promoting Iterative Joint NNLS (SPIJN) algorithm. Here, we compared both methods in several brain regions from healthy participants and normal-appearing and lesion tissue from MS patients (EDSS 0-1.5). We found that NNLS-based lesion-average MWF was rather comparable to white matter while SPIJN-based MWF was lower. Thus, SPIJN could be more sensitive to demyelination in lesions but comparisons to gold standard techniques are clearly needed.

Qian Chen¹, Jilei Zhang², and Bing Zhang^1,3
1Department of Radiology, Drum Tower Hospital, Clinical College of Nanjing Medical University, Nanjing, China, ²Philips Healthcare, Shanghai, China, ³Department of Radiology, Drum Tower Hospital, Medical School of Nanjing University, Nanjing, China

Cholinergic degeneration in the basal forebrain (BF) plays a crucial role in Alzheimer’s disease. In the present study, reduced BF volumes and damaged BF-associated tracts were observed in patients with mild cognitive impairment (MCI) compared to those with normal aging. Correlation and mediation analyses highlighted that volume reductions in the BF mainly accounted for deficits in global cognition, memory, and visuospatial ability, while disruption in BF-cortical tracts mainly contributed to executive and processing speed dysfunction. Collectively, this study may provide a better understanding of BF damage underlying cognitive impairment and provide guidance for intervention targets in MCI patients.