Content uploaded by Tobias Kober
Author content
All content in this area was uploaded by Tobias Kober on Oct 26, 2022
Content may be subject to copyright.
NeuroImage: Clinical 36 (2022) 103205
Available online 24 September 2022
2213-1582/© 2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Cortical lesions, central vein sign, and paramagnetic rim lesions in multiple
sclerosis: Emerging machine learning techniques and future avenues
Francesco La Rosa
a
,
b
,
c
,
*
, Maxence Wynen
b
,
d
,
e
,
f
, Omar Al-Louzi
g
,
h
, Erin S Beck
c
,
g
,
Till Huelnhagen
a
,
f
,
i
, Pietro Maggi
e
,
j
,
k
, Jean-Philippe Thiran
a
,
b
,
f
, Tobias Kober
a
,
f
,
i
, Russell
T Shinohara
l
,
m
,
n
, Pascal Sati
g
,
h
, Daniel S Reich
g
, Cristina Granziera
o
,
p
, Martina Absinta
q
,
r
,
Meritxell Bach Cuadra
b
,
f
a
Signal Processing Laboratory (LTS5), Ecole Polytechnique F´
ed´
erale de Lausanne (EPFL), Lausanne, Switzerland
b
CIBM Center for Biomedical Imaging, Switzerland
c
Department of Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, USA
d
ICTeam, UCLouvain, Louvain-la-Neuve, Belgium
e
Louvain Inammation Imaging Lab (NIL), Institute of Neuroscience (IoNS), UCLouvain, Brussels, Belgium
f
Radiology Department, Lausanne University and University Hospital, Switzerland
g
Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, USA
h
Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, USA
i
Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland
j
Department of Neurology, Cliniques universitaires Saint-Luc, Universit´
e catholique de Louvain, Brussels, Belgium
k
Department of Neurology, CHUV, Lausanne, Switzerland
l
Center for Biomedical Image Computing and Analysis (CBICA), Department of Radiology, University of Pennsylvania, Philadelphia, PA, USA
m
Penn Statistics in Imaging and Visualization Endeavor (PennSIVE), Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania, Philadelphia, PA,
USA
n
Department of Biostatistics, Epidemiology, and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA
o
Translational Imaging in Neurology (ThINk) Basel, Department of Biomedical Engineering, Faculty of Medicine, University Hospital Basel and University of Basel,
Switzerland
p
Neurologic Clinic and Policlinic, MS Center and Research Center for Clinical Neuroimmunology and Neuroscience Basel (RC2NB), University Hospital Basel and
University of Basel, Basel, Switzerland
q
IRCCS San Raffaele Hospital and Vita-Salute San Raffaele University, Milan, Italy
r
Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, USA
ABSTRACT
The current diagnostic criteria for multiple sclerosis (MS) lack specicity, and this may lead to misdiagnosis, which remains an issue in present-day clinical practice.
In addition, conventional biomarkers only moderately correlate with MS disease progression. Recently, some MS lesional imaging biomarkers such as cortical lesions
(CL), the central vein sign (CVS), and paramagnetic rim lesions (PRL), visible in specialized magnetic resonance imaging (MRI) sequences, have shown higher
specicity in differential diagnosis. Moreover, studies have shown that CL and PRL are potential prognostic biomarkers, the former correlating with cognitive
impairments and the latter with early disability progression. As machine learning-based methods have achieved extraordinary performance in the assessment of
conventional imaging biomarkers, such as white matter lesion segmentation, several automated or semi-automated methods have been proposed as well for CL, PRL,
and CVS. In the present review, we rst introduce these MS biomarkers and their imaging methods. Subsequently, we describe the corresponding machine learning-
based methods that were proposed to tackle these clinical questions, putting them into context with respect to the challenges they are facing, including non-
standardized MRI protocols, limited datasets, and moderate inter-rater variability. We conclude by presenting the current limitations that prevent their broader
deployment and suggesting future research directions.
Abbreviations: MS, multiple sclerosis; MRI, magnetic resonance imaging; DL, deep learning; ML, machine learning; CL, cortical lesions; PRL, paramagnetic rim
lesions; CVS, central vein sign; WML, white matter lesions; FLAIR, uid-attenuated inversion recovery; MPRAGE, magnetization prepared rapid gradient-echo; GM,
gray matter; WM, white matter; PSIR, phase-sensitive inversion recovery; DIR, double inversion recovery; MP2RAGE, magnetization-prepared 2 rapid gradient
echoes; SELs, Slowly evolving/expanding lesions; CNN, convolutional neural network; XAI, explainable AI; PV, partial volume.
* Corresponding author at: Department of Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
E-mail address: francesco.larosa@mssm.edu (F. La Rosa).
Contents lists available at ScienceDirect
NeuroImage: Clinical
journal homepage: www.elsevier.com/locate/ynicl
https://doi.org/10.1016/j.nicl.2022.103205
Received 13 December 2021; Received in revised form 9 September 2022; Accepted 16 September 2022
NeuroImage: Clinical 36 (2022) 103205
2
1. Introduction
Multiple sclerosis (MS) is a chronic inammatory disease and a
common cause of neurological disability in young adults (Reich et al.,
2018). Its hallmark is demyelinated white matter lesions (WML) forming
in the central nervous system (Reich et al., 2018). These lesions are
assessed in-vivo with magnetic resonance imaging (MRI), which is the
imaging technique of choice to diagnose MS and monitor the disease
over time (Hemond and Bakshi, 2018). The current MRI diagnostic
criteria (McDonald criteria) are based on the dissemination in space and
time of such lesions (Thompson et al., 2018). Moreover, the quanti-
cation of the total lesion volume is important to determine ongoing
disease activity and monitor treatment effect over time (Giorgio et al.,
2014). Recommended MRI techniques include T2 and T1-weighted
inversion recovery sequences, such as uid-attenuated inversion re-
covery (FLAIR), and magnetization prepared rapid gradient-echo
(MPRAGE) (Wattjes et al., 2021). At common clinical magnetic elds
(1.5 T and 3 T), the use of gadolinium-based contrast agents is useful to
evaluate patients suspected of MS and monitor disease activity causing
breakdown of the blood–brain barrier (Filippi et al., 2019).
As the manual detection of WML is time-consuming and prone to
inter-rater variability (Hagens et al., 2019), a myriad of automated or
semi-automated approaches have been developed to facilitate this task
(Llad´
o et al., 2012). These methods were initially based primarily on
MRI intensity features and probabilistic atlases (Llad´
o et al., 2012),
whereas, more recently, the vast majority use deep learning (DL) ap-
proaches (Zeng et al., 2020), without prior feature extraction. Sub-
stantial effort is now being made towards reproducibility of the results
and open science (Vrenken et al., 2021). Several grand challenges have
been organized (Carass et al., 2017; Commowick et al., 2018; Commo-
wick et al., 2021), in which DL-based methods have achieved the best
performance, approaching or sometimes even outperforming human
readers (Carass et al., 2017; Commowick et al., 2021). WML segmen-
tation methods have been reviewed recently (Zeng et al., 2020; Kaur
et al., 2021); the present review thus focuses on machine learning
techniques tailored for lesional biomarkers specic to MS that require
advanced MRI techniques and have the potential to improve MS diag-
nosis and prognosis.
One major drawback of the current MS diagnostic criteria is their
lack of specicity, as they were proposed to identify patients with a high
likelihood of MS rather than distinguish MS from other conditions
(Thompson et al., 2018). The lack of specicity of these criteria may lead
to misdiagnosis, which remains a persistent problem of MS (Solomon
et al., 2019). Multi-center studies have shown a misdiagnosis rate of
18% (Kaisey et al., 2019), often associated with atypical clinical or
imaging ndings. Improving the diagnostic specicity would prevent
harmful consequences for patients (Solomon et al., 2016) and allow
clinicians to prescribe the appropriate treatment earlier. In addition,
although clinical relapses are often associated with the appearance of
new WML, the overall WML burden, which is the most common MRI
biomarker examined in clinical routine, is only moderately correlated
with disability and poorly predicts transition to progressive disease
(Barkhof, 2002). For all these reasons, there is a need for additional
biomarkers that are highly specic to MS or correlate with disease
progression.
Quantitative MRI, such as relaxometry, myelin imaging, or diffusion
MR, provides information related to the microstructural composition
and organization of tissues. In MS, quantitative MRI techniques com-
plement conventional MRI techniques by providing insights into disease
mechanisms (Granziera et al., 2021). For instance, diffusion tensor im-
aging and microstructure models of diffusion can help better understand
the MS lesion heterogeneity (myelin and axonal damage). Voxel-wise
analysis methods allow exploring group-wise differences without the
need for prior lesion segmentation (Thaler et al., 2021;16(2):
e0245844.). On the contrary, classication methods in this context have
been used to cluster different lesion types based on prior lesion
segmentation and derived scalar measurements from diffusion-based
measurements (FA, MD, NODDI parameters, etc) averaged at the
lesion level (Lu et al., 2021; Oladosu et al., 2021; Ye et al., 2020; Mar-
tínez-Heras et al., 2020). Further studies, however, are still needed to
verify the possible use of these quantitative features for patient
stratication.
Recently, advances in MR technology, such as the development of
specialized sequences, acceleration of protocols, and the proliferation of
ultra-high eld MRI, have allowed the imaging of pathologically specic
MS lesional biomarkers (Cortese et al., 2019; Ineichen et al., 2021).
These include cortical lesions (CL), the central vein sign (CVS), and
paramagnetic rim lesions (PRL). Studies have shown that CL and PRL are
potential prognostic biomarkers: CL are associated with cognitive im-
pairments, while patients with PRL experience an earlier progression in
disability (Calabrese et al., 2010; Absinta et al., 2019). Furthermore, the
CVS and PRL have proven to be effective for differentiating MS from
mimicking diseases (Ontaneda et al., 2021; Sati et al., 2016; Clarke et al.,
2020; Maggi et al., 2020). All three biomarkers, however, require
dedicated MRI sequences at high (3 T) or ultra-high (7 T) magnetic
elds, and experienced raters for their manual assessment, which can be
very time-consuming. As done in the past for WML, various automated
or semi-automated methods, mostly based on machine learning (ML),
have been developed to facilitate the three aforementioned biomarkers’
assessment (see Table 1). Compared to their WML counterparts, how-
ever, they face additional challenges, including non-standardized im-
aging protocols, moderate inter-rater variability when determining
ground truth annotations, and smaller datasets. Automated assessment
could improve standardization and facilitate large-scale assessment in
clinical routine of the aforementioned biomarkers, with clear benets in
terms of MS diagnosis and prognosis.
In this review, we rst briey describe these advanced imaging
biomarkers and their imaging requirements and then focus on image
processing techniques tailored for their automated segmentation and
classication. We conclude with a discussion on current limitations and
future lines of research to boost the development of ML approaches in
this area and encourage their adoption in MS research and clinical
settings.
2. Cortical lesions, paramagnetic lesions, and central vein sign
In this section, we present a brief description of CL, CVS, and PRL,
and their respective imaging protocols. In addition to the CVS and PRL,
which have emerged as promising MS biomarkers in recent years, we
also included CL which, although included in the MS diagnostic criteria,
are not yet commonly analyzed in clinical practice. For the sake of
completeness, a short description of slowly expanding lesions (SELs) is
also provided, although these have not been assessed with ML-based
approaches yet.
Cortical lesions (CL) - Cortical lesions are a type of MS lesions that
involve, at least partially, the cortex and have been classied into three
main categories (Calabrese et al., 2010) (see Fig. 1): leukocortical le-
sions are located at the interface between WM and gray matter (GM)
(type I), intracortical lesions are purely in the cortex and do not reach
the pial surface (type II), and subpial lesions touch the subpial surface of
the brain (type III) and sometimes extend all the way to the white matter
(type IV). Cortical demyelination in MS has long been recognized in
pathology studies, but only in the last two decades have dedicated se-
quences on high- and ultra-high eld scanners provided in-vivo evidence
of cortical damage (Calabrese et al., 2010). Cortical lesions are clinically
interestingfor several reasons. First, they have been observed in the
early stages of the disease and in all of the major MS phenotypes (Kidd
et al., 1999). Second, they are associated with disability (Harrison et al.,
2015; Nielsen et al., 2013; Calabrese et al., 2012) and in some studies,
their number was associated with cognitive disability more strongly
than the number of WML (Harrison et al., 2015; Favaretto et al., 2016).
Third, longitudinal studies have linked them with disease progression
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
3
(Treaba et al., 2019; Mainero et al., 2015; Scalfari et al., 2018; Calabrese
et al., 2013). Fourth, subpial cortical demyelination is highly specic to
MS (Junker et al., 2020); CL have been observed in patients with
radiologically isolated syndrome (Giorgio et al., 2011), but not in pa-
tients with neuromyelitis optica (Sinnecker et al., 2012). Since 2017, CL
have been included in the MS diagnostic criteria (Thompson et al.,
2018), but their visualization from routine MRI sequences remains
difcult. For instance, a postmortem study showed that 3D FLAIR at 3 T
could detect about 41% of leukocortical lesions and only 5% of intra-
cortical and subpial lesions (Geurts et al., 2005). This supports the need
for specialized MRI techniques (see Fig. 2) such as the phase-sensitive
inversion recovery (PSIR), double inversion recovery (DIR), and
magnetization-prepared 2 rapid gradient echoes (MP2RAGE) (Filippi
et al., 2019; Müller et al., 2022). However, these sequences are still
relatively insensitive to CL at 1.5 T and 3 T (Müller et al., 2022; Kilsdonk
et al., 2016; Beck et al., 2020). Ultra-high eld MRI, with its higher
signal-to-noise ratio and increased susceptibility effects, has proven to
be a powerful tool for increasing the sensitivity to CL, especially for
intracortical and subpial lesions (Madsen et al., 2021; Beck et al., 2022;
Maranzano et al., 2019). Even with the most sensitive methods, how-
ever, CL are small and often subtle, making manual segmentation
extremely time consuming and subject to relatively low inter-rater
reliability (Harrison et al., 2015; Faizy et al., 2017).
Central vein sign (CVS) - Recently, studies have suggested that an
MRI-detectable central vein inside MS lesions might be evidence of
pathological processes specic to MS (see Fig. 3) (Maggi et al., 2018;
Solomon et al., 2016). This marker, referred to as the “central vein sign,”
has gained attention in recent years, as it could help to differentiate MS
from mimicking diseases (Sati et al., 2016; Sinnecker et al., 2019; Ciotti
et al., 2022; Sparacia et al., 2018; Tranfa et al., 2022). Small cerebral
veins can be detected with susceptibility-based MRI sequences, taking
advantage of the magnetic properties of venous blood that is rich in
deoxyhemoglobin (Haacke et al., 2009; Mittal et al., 2009). The CVS can
be reliably observed across different T2* sequences at 3 T, although the
sensitivity depends on the sequence considered (Samaraweera et al.,
2017). To obtain the best detection sensitivity for the CVS, optimized
MRI acquisitions have been proposed (T2*-weighted acquired with 3D-
segmented echo-planar-imaging or T2*w 3D-EPI (Sati et al., 2014),
combined T2-FLAIR and T2*, also called FLAIR* (Sati et al., 2012), and
susceptibility-based sequence, called SWAN-Venule (Gait´
an et al.,
2020). These sequences were shown to provide superior CVS detection
compared to clinical acquisitions at 1.5 T and 3 T (Castellaro et al., 2020;
Suh et al., 2019). Single-center and multi-center retrospective studies
imaging patients with clinically established diagnoses have demon-
strated a signicantly higher proportion of CVS-positive white matter
lesions (%CVS +) in MS (mean pooled incidence: 79%, 95% CI:
68–87%) (Suh et al., 2019) as compared to other neurological disorders
mimicking MS (mean pooled incidence: 38%, 95% CI: 18–63%) (Suh
et al., 2019) such as cerebral small vessel disease (Campion et al., 2017),
neuromyelitis optica spectrum disorder (NMOSD) (Cortese et al., 2018),
inammatory vasculopathies (Maggi et al., 2018), and migraine (Solo-
mon et al., 2016). To distinguish MS from other neurological conditions,
different CVS-based criteria have been proposed to date, some based on
the percentage of perivenular lesions (from 35% to 60%) and others
simply on the CVS lesion count (3-lesion or 6-lesion rule) (Maggi et al.,
2018; Tallantyre et al., 2011; Mistry et al., 2016; Solomon et al., 2018).
From a diagnostic perspective, retrospective studies have shown excel-
lent diagnostic discrimination by applying the ‘40% rule’ (Tallantyre
et al., 2011) with sensitivity =91% [95% CI, 82%-97%] and specicity
=96% [95% CI, 88%-100%]) (Castellaro et al., 2020). However,
applying percentage-based criteria requires manual exclusion of lesions
that are conuent or have multiple or eccentric veins, and performing
the CVS evaluation on all the remaining lesions present in patients’
brains, which is a time-consuming process difcult to accomplish in
clinical practice.
Paramagnetic rim lesions (PRL) - Recent pathology studies have
Table 1
Summary of the methods proposed for the automated or semi-automated analysis of cortical lesions, the central vein sign, and paramagnetic rim lesions. The task is
abbreviated as follows: segmentation (S), classication (C). If not specied, all sequences were 3D. Other abbreviations: k-nearest neighbors algorithm (K-NN),
convolutional neural network (CNN), partial volume (PV).
Biomarker Authors (year) Method Task MRI sequences (magnetic eld
strength)
Dataset size
(n. of sites)
Code
available
Cortical
lesions
Tardif,C. L., et al. (Tardif et al., 2010)
(2010)
Laminar prole shape analysis S Quantitative high-resolution
scan (3 T)
1 post mortem brain
scan (1)
No
Fartaria, M.J., et al. (Fartaria et al.,
2016) (2016)
K-NN S FLAIR, MPRAGE, MP2RAGE,
DIR (3 T)
39 MS patients (1) No
Fartaria, M.J., et al. (Fartaria et al.,
2017) (2017)
K-NN with partial volume
constraints
S FLAIR, MPRAGE, MP2RAGE,
DIR (3 T)
39 MS patients (1) No
Fartaria, M.J., et al. (Fartaria et al.,
2019)(2019)
PV estimation and topological
constraints
S MP2RAGE (7 T) 25 MS patients (2) No
La Rosa, F., et al. (La Rosa et al., 2020)
(2020)
CNN S FLAIR, MP2RAGE (3T) 90 MS patients (2) Yes
b
La Rosa, F., et al. (La Rosa et al., 2020)
(2020)
CNN S and
C
MP2RAGE, 2D T2*-w GRE,
T2*w 3D-EPI (7T)
60 MS patients (1) Yes
c
La Rosa, F., et al
(2022) (La Rosa et al., 2022)
CNN S
and
C
MP2RAGE, 2D T2*-w GRE,
T2*w 3D-EPI (7T)
80 MS patients (2) Yes
c
Paramagnetic rim
lesions
Barquero, G., et al. (Barquero et al.,
2020) (2020)
CNN C FLAIR, T2*w 3D-EPI (3T) 124 MS patients (2) No
Lou, C., et al. (Lou et al., 2021) (2021) Random forest classier C FLAIR, T1-w, T2*w 3D-EPI (3T) 20 MS patients (1) Yes
d
Zhang, H. et al. (2022) (Zhang et al.,
2022)
Residual network and
radiomic features
C FLAIR, QSM (3T) 172 MS patients (1) No
Central vein sign Maggi, P., Fartaria, MJ, et al. (Maggi
et al., 2020) (2020)
CNN C T2*w 3D-EPI, FLAIR (3T) 42 MS patients, 33 mimics,
5 others (3)
No
Dworkin, J. D., et al. (Dworkin et al.,
2018)
(2018)
Probabilistic method C T2*w 3D-EPI, FLAIR (3T) 16 MS patients, 15 MS
mimics (1)
Yes
e
b
https://github.com/FrancescoLR/MS-lesion-segmentation
c
https://github.com/Medical-Image-Analysis-Laboratory/CLaiMS
d
https://github.com/carolynlou/prlr
e
https://github.com/jdwor/cvs
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
4
demonstrated that about 30% of chronic demyelinated lesions are
pathologically characterized by perilesional accumulation of iron-laden
microglia and macrophages, showing evidence of smouldering demye-
lination and axonal loss around an inactive hypocellular core (see Fig. 4)
(Frischer et al., 2015; Luchetti et al., 2018). This type of MS lesion has
been dened as “chronic active/smouldering lesions”. Due to their pe-
ripheral paramagnetic iron rim, these lesions can be depicted using in-
vivo susceptibility-based MRI techniques (T2*-weighted magnitude,
phase images, and quantitative susceptibility mapping, QSM) at both 3 T
and 7 T (Absinta et al., 2018; Absinta et al., 2016), and are therefore
termed “paramagnetic rim lesions” (PRL).
Direct comparison among different MRI sequences and post-
processing techniques for PRL detection is still limited. A recent study
(Huang et al., 2022) has compared QSM and high-pass-ltered (HPF)
phase imaging for identifying PRL. Of 2062 MS lesions detected in 80
patients, 9.1% were identied as PRL in both QSM and HPF phase, 9.8%
were PRL only in HPF phase, and the rest were rim negative. QSM-
identied PRL showed stronger association with clinical disability
compared to those detected by HPF phase imaging.
Overall, in vivo studies have shown that about 50% of relapsing and
about 60% of progressive MS patients have at least one PRL (Absinta
et al., 2019; Maggi et al., 2020). Of clinical relevance, PRL accrual has
been recently linked to a more aggressive disease course and disability
accumulation at a younger age and/or shorter disease duration (Absinta
et al., 2019). Reasons for such association directly rely on a few typical
features of these lesions: PRL are destructive (Absinta et al., 2016; Kolb
et al., 2021), they do not remyelinate (Absinta et al., 2016), and they can
expand over time, (Absinta et al., 2019) demyelinating the surrounding
tissue and injuring axons, as corroborated by the elevation of serum
neurolament light chain in patients with PRL who are not forming new
white matter lesions (Maggi et al., 2021). The recent discovery that the
paramagnetic rim can signicantly shrink or disappear (Absinta et al.,
2021; Dal-Bianco et al., 2021) holds promise regarding its potential use
as an outcome measure in clinical trials designed to halt the chronic
inammation at the lesion edge. In addition to their prognostic role, PRL
appear specic to MS, as they have been rarely detected in patients with
other neurological conditions (52% of MS vs 7% of non-MS in a multi-
center study of 438 individuals) (Maggi et al., 2020). PRL have the
promise of becoming a clinically relevant biomarker to both improve MS
diagnosis and monitor treatment efcacy over time.
Overall, there are not yet imaging guidelines for the visual detection
of PRL which requires specic training and remains challenging and
time-consuming. The development of ML-based approaches, described
in the next section, may help alleviate these issues and facilitate PRL
assessment.
Slowly evolving/expanding lesions (SELs) – A different computa-
tional approach, designed to detect in vivo longitudinal volumetric
lesional changes not associated with gadolinium enhancement, iden-
ties the so-called “slowly evolving/expanding lesions” or SELs. Linear
and radial lesion expansion is computed as a function of the Jacobian
determinant of the non-linear deformation eld between baseline and
follow up scans (linearity assessment requires a minimum of 3 scans)
(Elliott et al., 2019). Advantages of this approach relate to the use of
retrospective conventional T1-weighted and T2-weighted scans. re-
analysis of the ORATORIO
a
clinical trial found reduced rate of T1-
SELs expansion in progressive patients treated with ocrelizumab vs
placebo (Elliott et al., 2019). A recent study showed that SELs are in-
dependent predictors of EDSS worsening after a median follow up of 9
years (Preziosa et al., 2022). The neuropathological correlate of SELs is
currently not yet determined and preliminary data showed only modest
correlation with PRL (Elliott et al., 2021).
Overall, CL, PRL, and CVS have the potential to considerably
Fig. 1. Representative examples of the three main types of CL. From left to right: 3 T MP2RAGE (0.75 mm isometric), 7 T MP2RAGE (0.5 mm isometric), 7 T T2*-EPI
(0.5 mm isometric) and 7 T T2*-GRE (0.5 mm isometric). CL, including leukocortical, intracortical, and subpial subtypes, are seen better at 7 T due to higher signal-
to-noise ratios, allowing higher resolution scans, and increased susceptibility effects. The 7 T MP2RAGE image shown was obtained as the average of 4 acquisitions.
a
A phase 3, randomized, parallel-group, double-blind, placebo-controlled
trial.
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
5
improve the specicity of MS diagnosis (Junker et al., 2020; Maggi et al.,
2018; Maggi et al., 2018). Moreover, studies have shown that CL, PRL,
and SELs can be useful to assess prognosis (Calabrese et al., 2012;
Absinta et al., 2016). Their manual assessment, however, particularly
for CL, is both time-consuming and prone to inter-rater variability. As for
conventional WML, some automated or semi-automated methods have
been proposed to accelerate this task (Fartaria et al., 2017; Fartaria
et al., 2019; La Rosa et al., 2020; La Rosa et al., 2020; Fartaria et al.,
2016; Tardif et al., 2010; Barquero et al., 2020; Lou et al., 2021; Maggi
et al., 2020; Dworkin et al., 2018). In the next section, we describe the
challenges these approaches have been facing and how these differ from
the segmentation of WML.
2.1. Added challenges for CL, PRL, and CVS assessment
Compared to conventional imaging biomarkers, the visual
Fig. 2. examples of CL seen in different MRI contrasts at 3 T. From left to right: MP2RAGE, DIR, PSIR, IR-SWIET, T2*. Red arrows point to leukocortical lesions and
blue arrows to subpial lesions. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
Fig. 3. A central vein running through a lesion visible in the three planes (zoomed-in boxes) in a 3D FLAIR* obtained combining FLAIR and T2*-EPI acquisitions at 3
T. Resampling was applied to the magnied images for visualization purposes. FLAIR, T2*-EPI and FLAIR* are the MRI contrasts that have been used by ML ap-
proaches for CVS detection. Refer to the supplementary material for additional examples of the CVS on different susceptibility-weighted imaging sequences.
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
6
assessment of CL, PRL and CVS present some additional challenges.
Imaging and assessment guidelines- The rst obstacle is repre-
sented by the lack of consensus guidelines for imaging protocols.
Although efforts have been made to standardize the use of MRI in
clinical practice for conventional biomarkers (Wattjes et al., 2021),
guidelines are still in a preliminary stage for CL, PRL, and the CVS. CL
were included in the MS diagnostic criteria in 2017 (Thompson et al.,
2018), but, currently, there is no single gold standard sequence at 3 T for
their detection in a clinical setting. PSIR, DIR, and MP2RAGE are all
recommended by an international consensus (Filippi et al., 2019).
However, these contrasts remain primarily acquired in research settings
and are not yet widely used in clinical routine. Moreover, although 7 T
MRI is increasingly used to detect CL, no guidelines have been presented
yet to standardize their imaging sequences and their identication.
Regarding the CVS, in a 2016 consensus statement, the North
American Imaging in MS Cooperative (NAIMS) proposed a standard
radiological denition and suggested specic MRI acquisitions (Sati
et al., 2016). Following these recommendations, recent studies have
shown that high-resolution T2*w 3D-EPI or FLAIR* improve the detec-
tion of the CVS compared to clinical acquisitions (Castellaro et al., 2020;
Suh et al., 2019). Nevertheless, a standardized clinical protocol for CVS
detection is still missing. Among the three aforementioned biomarkers,
PRL is probably the one at the earliest stages. Although recent studies
support the feasibility of its assessment on clinical scans and its utility in
improving the diagnosis and prognosis of MS (Maggi et al., 2020), there
are currently no international guidelines for its denition nor a stan-
dardized MRI protocol for its analysis. Several different imaging mo-
dalities have been used for the PRL assessment, including phase 3D-EPI,
susceptibility weighted imaging (SWI), QSM, and multi-echo T2* GRE at
both 3 T and 7 T (Absinta et al., 2018; Absinta et al., 2016). However,
there is a paucity of studies that have systematically compared the
sensitivity of these acquisition techniques for PRL detection, especially
when implemented at different eld strengths.
These evolving or unclear criteria for CL, the CVS, and PRL, wide
variety of imaging settings, and lack of clear guidelines for standardized
protocols clearly jeopardize the development and wide use of these
biomarkers and of targeted ML techniques.
Expert assessment - Even for experts, the task of segmenting CL,
detecting the CVS, or classifying PRL is intrinsically more challenging
than segmenting WML. CL are generally smaller in size and more
affected by partial volume (PV) effects, compared to WML. The cortex is
convoluted, so lesion shape is not as regular as in WM, and traditional
methods of radiological evaluation (scrolling through an image stack)
are less effective in this context. The detection of the CVS requires
susceptibility-based MRI and its exclusion criteria need to be carefully
considered when performing its assessment (Sati et al., 2016).
Susceptibility-based images used to detect PRL present variability in the
susceptibility signal and several artifacts, therefore experienced raters
are needed. Moreover, as these three biomarkers have been so far mainly
studied in research settings, clinicians do not commonly see them in
clinical practice and might need specic training and dedicated time to
perform a proper assessment.
2.2. Machine learning specic challenges
From a ML perspective, the automated segmentation or classication
of CL, PRL, and the CVS faces new challenges as compared to their WML
counterparts.
Limited datasets - An additional limitation, particularly for super-
vised DL-based approaches, is the scarcity and limited size of datasets in
which these biomarkers were manually annotated. For their assessment,
CL, CVS, and PRL all require advanced MRI sequences at high or ultra-
Fig. 4. (A) Representative paramagnetic rim lesion seen on a 3 T T2*-weighted seg-EPI magnitude and unwrapped ltered phase in the three orthogonal planes
(zoomed-in red boxes, the rim is indicated with red arrows). The central vein (yellow arrows) is also clearly visible within the lesion. (B) Representative periven-
tricular MS lesion with a paramagnetic rim. Paramagnetic rims are visible on both unwrapped phase and QSM-reconstructed images (white arrows). (For inter-
pretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
7
high magnetic eld and experienced raters, and this makes it difcult to
have large multi-site datasets. Although national MS registries exist in
most countries, the data sharing of MRI in MS is still limited and often
includes only conventional sequences (Vrenken et al., 2021). Moreover,
the CVS or the rim-shape in PRL are visible only on a few slices per
lesion, reducing, even more, the data available to train a supervised
approach.
Inter-rater variability - The lack of standardization for both the
denition and imaging of these biomarkers contributes to a modest
inter-rater variability. Barquero et al. (Barquero et al., 2020) showed
that, in a cohort of 124 MS patients, approximately 38% of PRL needed a
consensus review from two raters classifying PRL independently (Cohen
k of 0.73). Absinta et al. observed similar inter-rater agreement between
three experts at 3 T (Fleiss coefcient of 0.71), with somewhat higher
intra-rater reliability (Cohen k of 0.77) (Absinta et al., 2018). Similar
values were reported at 7 T for the same set of patients, whereas the
agreement between 3 T and 7 T annotations was substantial (Cohen k of
0.78). In a similar way, the inter-rater agreement was shown to be
moderate for the segmentation of CL (Harrison et al., 2015; Nielsen
et al., 2012; Geurts et al., 2011) and high, but not perfect, for the CVS
(Cohen k of 0.9) (Maggi et al., 2020; Kau et al., 2013). Imaging quality
and motion artifacts are other factors to consider as they can result in
inconspicuity of all three biomarkers and, therefore, contribute to poor
inter-rater agreement. Overall, the inter-rater variability represents an
additional challenge for the development of automated approaches, as
there might be large inconsistencies in the annotations of the training or
testing set due to different raters performing the manual assessment.
3. Methods
Despite the recent discovery of the CVS and PRL and the above-
mentioned challenges, a few groups have already attempted to sup-
port their analysis with automated or semi-automated ML methods. To
these two novel biomarkers, we add also CL, which, although studied for
several years, have only recently been assessed automatically. As there
are no ML-based approaches to assess SELs yet, the prospect of analyzing
this additional biomarker with ML is presented in the Discussion section.
Overall, many fewer methods have been proposed for the assessment of
CL, PRL, and the CVS compared to WML. In what follows, we briey
describe these state-of-the-art techniques by grouping them according to
the biomarker they assess. A summary of the main characteristics for
each method is presented in Table 1, and a scheme of the MRI sequences
used to detect these three biomarkers at both 3 T and 7 T is shown in
Fig. 5.
3.1. Cortical lesions
ML-based methods automatically segmenting CL have been explored
with both 3 T and 7 T MRI. The rst work (Tardif et al., 2010) present in
the literature considered a postmortem MS brain imaged at 3 T with
different sequences (T1, T2, and relative proton density) at high reso-
lution (0.35 mm isotropic) (Tardif et al. (2012)). Tardif et al. (Tardif
et al., 2010) proposed to rst identify the cortical and white matter
surfaces, then extract laminar proles between the two tissues, and
nally apply a k-means classier to the prole intensity and shape fea-
tures to parcellate the cortex and detect lesions. Although showing
promising results on one postmortem MS brain, this method was never
validated with larger cohorts nor in-vivo data. A few years later, Fartaria
et al. (Fartaria et al., 2016) proposed the rst automated method for the
segmentation of both WM and cortical lesions. In their study, they
analyzed a cohort of 39 early-stage MS patients and considered both
conventional (FLAIR, MPRAGE) and advanced (MP2RAGE, DIR) MRI
sequences at 3 T. In a nutshell, their method consisted of co-registering
the different MRI contrasts, leveraging prior tissue probability maps
from existing brain atlases of healthy subjects, and nally classifying
each voxel either as being a lesion or healthy tissue with a k-nearest
neighbor (k-NN) algorithm. Additionally, as post-processing, all lesions
smaller than 3.6 µL were discarded, and a region-growing algorithm was
applied to improve the lesion delineation. Results were promising,
showing a CL detection rate of 62% when advanced imaging (FLAIR,
Fig. 5. Scheme showing the main MRI sequences used for detecting each biomarker at both 3 T and 7 T.
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
8
MP2RAGE, and DIR) was included. An extension of this segmentation
framework with a Bayesian partial volume (PV) estimation method was
presented by the same authors (Fartaria et al., 2017). They argued that
CL, being generally small and located at the interface between WM and
GM, suffer from strong PV effects. The addition of this PV model indeed
improved the delineation of CL in terms of both total lesion volume
estimation and dice coefcient (Fartaria et al., 2017).
The same research group also proposed a different segmentation
method for WML and CL using only 7 T MP2RAGE images (called
MSLAST: Multiple Sclerosis Lesion Analysis at Seven Tesla) (Fartaria
et al., 2019). MSLAST computes tissue concentration maps with a PV
algorithm and unies them based on topological constraints. A
connected-components analysis is then performed on gray matter and
cerebrospinal uid maps, and small components are classied as MS
lesions. This method was evaluated with 25 MS patients’ scans from two
research centers and reached a 58% patient-wise CL detection rate
(when 6
μ
L was considered as minimum lesion volume) with a false
positive rate of 40%. Moreover, it showed promising scan-rescan
repeatability within the same session, with a mean total lesion volume
difference (WML and CL combined) of 0.29 mL (mean total lesion vol-
ume 5.52 mL), vs 0.13 mL for the manual segmentations. More recently,
DL-based approaches have been presented as well (La Rosa et al., 2020;
La Rosa et al., 2020). In the rst study, La Rosa et al. proposed a
framework for the automated segmentation of WML and CL at 3 T using
FLAIR and MP2RAGE (La Rosa et al., 2020). Their method extracts 3D
patches of 88x88x88 voxels from the two MRI contrasts and feeds them
to a convolutional neural network (CNN). The CNN, inspired by the U-
Net, has an encoder and decoder path, each one with three resolution
levels. Evaluated on two datasets acquired in different centers, for a total
of 90 MS patients, the framework showed competitive performance,
with a CL detection rate of 76% and a false positive rate of 29%.
In a second study, the same group proposed a similar approach, this
time tailored exclusively for the detection of CL using multi-contrast 7 T
MRI (La Rosa et al., 2020). The contrasts considered were MP2RAGE,
T2*-weighted GRE, and T2*-w 3D-EPI. A cohort of 60 patients was
analyzed with a total of over 2000CL manually segmented by two ex-
perts. The CNN architecture proposed was similar to the one just
described, but with a modied output. In addition to the CL segmenta-
tion, the CNN provided a classication into two types (leukocortical and
intracortical/subpial lesions) and a separate branch with a simple tissue
segmentation in WM/GM. CL were correctly classied into the two types
by the network with an accuracy of 86%. Setting a minimum lesion size
of 0.75
μ
L, it achieved a CL detection rate of 67% with, however, a quite
high false positive rate of 42% (see Fig. 6). Importantly, about 24% of
these false positives were retrospectively judged as CL or possible CL by
an expert (La Rosa et al., 2020).
In a following publication (La Rosa et al., 2022), this method was
further improved and evaluated on a multi-site dataset. Its main modi-
cations included an added resolution level in the CNN architecture, a
larger 3D patch input size of 96x96x96 voxels, and the use of the focal
loss for training. Finally, a domain adaptation approach was applied to
verify the performance on external datasets. On 20 MRI scans of patients
imaged in a different center, this method achieved superior performance
(CL detection rate of 71%) compared to MSLAST (48%) when setting a
minimum lesion size of 6
μ
L.
3.2. The central vein sign
As of today, two automated ML methods for the classication of MS
lesions as CVS+(MS lesions showing the presence of the CVS) or CVS-
(MS lesions without the CVS) have been proposed in the literature
(Maggi et al., 2020; Dworkin et al., 2018). Both approaches were
developed and evaluated only with 3 T MRI. Dworkin et al. (Dworkin
et al., 2018) proposed a probabilistic method based on the Frangi ves-
selness lter (Frangi et al., 1496). They rst perform an automated WML
segmentation using T1 and FLAIR 3D MRI volumes acquired at 3 T
(Valcarcel et al., 2018; Valcarcel et al., 2018) and obtain a map of the
veins by applying the vesselness lter to a T2*w 3D EPI image. Conuent
lesions are then separated, and lesion centers are detected by textural
analysis (Dworkin et al., 2019). Periventricular lesions are discarded as
suggested by consensus guidelines (Sati et al., 2016), and a permutation
algorithm is applied to verify whether veins occur at the lesions’ centers
more often than would be expected due to random chance. Finally, to
account for scan motion, the single lesion CVS +probabilities are
weighted by the noise in their T2*-w 3D-EPI intensities and averaging
across the total number of lesions, a patient-wise CVS value is obtained.
This method was evaluated on a cohort of 31 adults, of whom 16 had
MS. When considering a 40% cutoff rule, the method yielded a sensi-
tivity of 0.94 and a specicity of 0.67 on a patient-wise classication
level. The performance of the method on a lesion-wise level was not
assessed. Although still far from experts’ performance, this was a rst
attempt to automatize the CVS assessment and encouraged further
improvements.
Maggi, Fartaria et al. (Maggi et al., 2020) introduced an optimized
CNN for the automated CVS assessment, called CVSnet. CVSnet is
inspired by the VGGnet (Simonyan and Zisserman, 2015) but composed
of only three convolutional layers followed by rectied linear unit
(ReLU) activations. Dropout was applied in each layer, and then two
fully-connected layers of size 32 and 2, respectively, were added to
provide the output. The authors selected 3D patches of size 21x21x21
voxels as input for the network, where each patch was centered on an MS
lesion and FLAIR* was the only MRI contrast used. Moreover, an
ensemble of 10 networks with the same architecture was trained and the
probability outputs were averaged to provide the nal prediction. This
study considered a cohort of 80 patients imaged at three different sites,
of whom 42 had MS, 35 an MS-mimic, and 5 an unknown diagnosis. On
the test set, CVSnet reached a lesion-wise sensitivity, specicity, and
accuracy of 0.83, 0.75, and 0.79, respectively. On a patient-wise level,
Fig. 6. Three representative axial slices from one MS patient showing the CL
segmentation results of an automated CL segmentation method (La Rosa et al.,
2020). 7 T MP2RAGE (left column) and CL mask (right column) showing true
positives (green), false negatives (red), and false positives (blue) of the auto-
mated approach with respect to the CL manual segmentation. (For interpreta-
tion of the references to colour in this gure legend, the reader is referred to the
web version of this article.)
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
9
using a 50% cut-off, CVSnet achieved a sensitivity, specicity, and ac-
curacy of 0.89, 0.92, and 0.90, respectively, outperforming the vessel-
ness lter (Frangi et al., 1496) and approaching expert performance.
However, as argued by the authors, these results are not directly com-
parable with those of Dworkin at al. (Dworkin et al., 2018), as the
CVSnet considered different exclusion criteria to pre/select the lesions,
and the initial lesion segmentation was performed manually.
3.3. Paramagnetic rim lesions
To our knowledge, only three methods have been proposed so far for
the detection of rim-like features and classication of PRL (Barquero
et al., 2020; Lou et al., 2021; Zhang et al., 2022). All three methods
considered 3 T MRI sequences, whereas 7 T imaging has not yet been
explored for the automated assessment of PRL. Barquero et al. (2020)
introduced a DL-based approach (called RimNet) for the semi-
automated classication of PRL, which considered 3D FLAIR and
T2*w 3D-EPI and phase 3D-EPI images. RimNet’s architecture is inspired
by the VGGnet (Simonyan and Zisserman, 2015) and composed of two
parallel CNN (one for either FLAIR or T2*w 3D-EPI image and one for
the phase 3D-EPI image), where each CNN is made of three convolu-
tional layers followed by a max-pooling operation. 3D patches of size
28x28x28 (centered around each MS lesion) are fed to each branch, and
both high-level and low-level feature maps are concatenated. An auto-
mated lesion segmentation based on FLAIR and MPRAGE/MP2RAGE (La
Rosa et al., 2020; La Rosa et al., 2019) was modied by an expert to split
conuent lesions. The performance of RimNet was assessed on a cohort
of 124 adults with MS who underwent 3 T MRI at two different sites with
two scanners from the same vendor. Two experts annotated PRL inde-
pendently and reached consensus in a joint session (462 PRL in total).
The proposed multimodal approach based on FLAIR and phase 3D-EPI
image achieves lesion-wise sensitivity and specicity of 0.70 and 0.95,
respectively. When considering a previously identied clinical threshold
of 4 PRL (Oladosu et al., 2021) for classifying patients as “chronic
active” and “non-chronic active”, RimNet reaches an accuracy of 0.90
and an F1-score of 0.84. These values are within 5% of the single ex-
perts’ metrics, suggesting that RimNet could be a valuable tool in sup-
porting the PRL analysis. The main drawback of RimNet, however, is
that the method is not fully automated, as conuent lesions were split
manually by an expert.
Lou et al. (Lou et al., 2021), on the other hand, proposed a fully
automated ML method for PRL assessment. They considered a cohort of
20 subjects with MS imaged with 3D FLAIR, 3D MPRAGE, and T2*-w
3D-EPI and phase 3D-EPI images. One neurologist inspected the T2*
magnitude and unwrapped phase images and annotated PRL (113 PRL
over the entire cohort). The automated pipeline, after some pre-
processing steps that included lesion segmentation (Valcarcel et al.,
2018; Valcarcel et al., 2018), lesion center detection (Dworkin et al.,
2019), and lesion labeling, consisted of extracting 44 different lesion-
wise radiomic features. A random forest classier was then tted on
these features, and its ability to classify PRL was evaluated on a test set
of 4 patients. Sensitivity and specicity of 0.75 and 0.81, respectively,
were achieved. Although fully automated, this study has three limita-
tions. First, the extremely small testing dataset (4 patients only with 47
PRL), annotated by a single expert, does not guarantee the generaliza-
tion of the proposed method. Second, all patients analyzed had at least
one PRL, and this might add a bias to the trained model. Finally, as
acknowledged by the authors, about 65% of misclassied lesions were
conuent, highlighting the need for a better solution to address these
lesions.
Inspired by these two previous works, Zhang et al. introduced QSM-
RimNet (Zhang et al., 2022), a QSM-based approach that combines a
two-branch feature extraction network and a synthetic minority over-
sampling technique. QSM-RimNet receives as input 3D patches of size
32x32x16 voxels where a masking out of non-lesional area is applied.
One branch of the network employs residual blocks to extract
convolutional features from QSM and FLAIR images, whereas the second
consists in a fully-connected network that processes previously obtained
radiomic features. Convolutional and radiomic features are concate-
nated and a minority oversampling network is used to alleviate the issue
of class imbalance. Finally, a probability of being a PRL is assigned to
each lesion. QSM-RimNet was evaluated with a stratied 5-folds cross-
validation over 172 MS patients with a total of 177 PRL. Compared to
RimNet and the automated approach of Lou et al., it outperformed both
methods achieving a lesion-wise sensitivity and specicity of 0.68 and
0.99, respectively, although the differences were not statistically sig-
nicant. Ablation studies showed that fusing convolutional and radio-
mic features improves the PRL identication (Zhang et al., 2022). Of
note, QSM-Rimnet is not fully-automated as during training and evalu-
ation it relies on manual corrections by experts of both PRL and
conuent lesions. Similarly to RimNet, this strong limitation currently
prevents its wider deployment and applicability.
Overall, two methods have tackled the PRL detection problem
considering mainly the T2*-w 3D-EPI sequence and one method has
focused on the QSM. Thus, none of the three frameworks has investi-
gated the effect of differences in SWI and QSM processing on ML-based
tools performance and this important aspect should be explored in
future studies.
4. Discussion
The methods described in the present review tackle challenging and
clinically relevant problems. Automated and reliable solutions for
detecting, classifying, and segmenting CL, PRL, and CVS are needed to
improve the standardization of these biomarkers and facilitate their
assessment in clinical routine. As of today, however, these methods are
still in an early stage and are slightly less sensitive than WML segmen-
tation approaches.
Nevertheless, such tools would provide obvious advantages, either as
stand-alone or adjunctive approaches as all three biomarkers are dif-
cult and time-consuming to analyze using conventional radiological
workows. In these particular cases, manual reading is so involved that
automated methods might actually boost the biomarkers’ widespread
adoption. First, they can substantially reduce analysis time, as compared
to a manual rating. Maggi, Fartaria et al., for instance, showed that
CVSNet was 600-fold faster on the test set compared to the manual
assessment (4 s vs 40 min) when considering a 50% CVS +lesions
criteria to distinguish MS from MS mimics (Maggi et al., 2020). A lower
time gain, however, would be expected if CVS +lesion-count criteria,
such as the 3-lesion and 6-lesion, were to be considered. Reduced
analysis time can be predicted also for PRL and CL assessment. In La
Rosa et al. (La Rosa et al., 2020), for instance, the automated CL seg-
mentation of one subject is computed on average in 20 s. Although a
direct comparison with the manual labeling was not reported, seg-
menting CL manually is known to be a much more time-consuming
process. A second main advantage of automated ML methods is their
ability to base their decision on 3D multi-contrast MRI analyzed
simultaneously. This stands in contrast to expert reviews, which typi-
cally involve comparison of 2D slices across several contrast mecha-
nisms in a variety of planes and are thus inherently limited in the
amount of information that can be readily gleaned.
4.1. Common trends
Some common trends can be observed in most of the proposed
pipelines. The large majority of the methods are supervised, relying on
expert annotations. Regarding the DL-based approaches, they all used
patch-based 3D CNN, exploiting the 3D intrinsic information, and often
considered more than one MRI contrast simultaneously. In addition, a
shared tendency consists of the use of relatively shallow architectures,
with a limited number of trainable parameters, due to the lack of large
datasets (La Rosa et al., 2020; La Rosa et al., 2020; Barquero et al., 2020;
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
10
Maggi et al., 2020). Combining this with extensive data augmentation
techniques can help when datasets are small and unbalanced. Alterna-
tively, other groups have tackled the problem of overtting by proposing
approaches based on classical ML techniques, such as k-NN (Fartaria
et al., 2017; Fartaria et al., 2016) or random forest classier (Commo-
wick et al., 2018). In these studies, either intensity-based, radiomic, or
probabilistic features are extracted and then fed to the respective clas-
sier. Overall, their current performance is inferior compared to their
DL-based counterparts.
In addition, some common pre-processing steps can also be identi-
ed. First, some methods use intensity normalization techniques, either
based on entire 3D volumes (Lou et al., 2021; Dworkin et al., 2018;
Fartaria et al., 2017; Fartaria et al., 2019; La Rosa et al., 2020; La Rosa
et al., 2020) or on single patches (Barquero et al., 2020; Maggi et al.,
2020). Second, the approaches using multiple MRI contrasts always
register all images to the same space (Lou et al., 2021; Fartaria et al.,
2017; Fartaria et al., 2019; La Rosa et al., 2020; La Rosa et al., 2020).
Registration errors might affect the methods’ performance. Finally, a
shared pre-processing step in all approaches for the CVS or PRL assess-
ment is the prior WML segmentation, obtained either manually (Maggi
et al., 2020) or with an automated tool (Barquero et al., 2020; Lou et al.,
2021; Dworkin et al., 2018). In both cases, this can be a source of error
that negatively affects the subsequent biomarkers’ classication
accuracy.
4.2. Current limitations
Currently, a major limitation hinders the deployment of the above-
described methods to the clinic: the methods proposed were trained
and evaluated on small datasets acquired from one or at most two
centers. Moreover, the MRI protocols used were often similar and not
representative of the current diversity of images acquired in the clinics,
including different processing, scans affected by noise and artifacts or
protocols missing certain modalities. Therefore, the automated ML
methods’ robustness on larger datasets and different scanners, especially
from multiple vendors, remains to be proven. This limitation is
emphasized by the current lack of standardized acquisition protocols
which increases the diversity of the MRI sequences considered for the
same biomarkers. This also represents a major hurdle for potential
regulatory approval of such methods. As regulatory approval is neces-
sary for widespread adoption in the clinics, which is, in turn, the pre-
requisite for the availability of large datasets, this is currently a circular
dependency issue.
In addition, the achieved performance levels of the automated ML
methods are still inferior compared to the human experts. Considering
the high inter-rater variability and the limited amount of data available,
there is also a considerable risk of having methods that perform well on
data annotated by a single expert and not as well with annotations from
other raters. To mitigate this issue, several methods have already
considered consensus annotations from two or more experts (La Rosa
et al., 2020; Barquero et al., 2020; Maggi et al., 2020). Regarding CL, no
automated method presented in the literature was compared, on the
same dataset, with the experts’ inter-rater variability, thus a proper
evaluation is not possible. With respect to CVS, Maggi, Fartaria et al.
(Maggi et al., 2020) compared the performance of CVSnet with the
consensus of two experts. Following the “50% rule,” CVSnet achieved on
the testing set a classication accuracy of 79%, whereas the experts
reached 100% accuracy in differentiating MS and mimic diseases. In a
similar way, Barquero et al. (Barquero et al., 2020) compared RimNet’s
performance with those of two experts in classifying PRL. In a lesion-
wise analysis, RimNet achieved a sensitivity of 71% and a negative
predictive value of 96%, approaching the experts, who reached 78% and
98%, respectively.
Another main limitation is represented by the fact that some methods
presented are not fully automated. CVSnet (Maggi et al., 2020), for
instance, used manually annotated MS lesion masks in which lesions
were excluded based on the NAIMS criteria (Sati et al., 2016), whereas
in the pipeline proposed by Dworkin et al. (Dworkin et al., 2018), scans
affected by noise were discarded following a manual rating. Similarly,
RimNet (Barquero et al., 2020) exploits lesion masks where conuent
lesions have been previously split into single units by an expert. In
contrast, all methods described to date for CL segmentation or detection
are fully automated (Fartaria et al., 2017; La Rosa et al., 2020; La Rosa
et al., 2020; Fartaria et al., 2016). Another persistent issue in the auto-
mated analysis of the CVS and PRL is the presence of conuent lesions.
Large, periventricular white matter lesions which include several single
units pose additional challenges as the current methods classify each
lesion singularly (Lou et al., 2021; Dworkin et al., 2018), and some of
them extract 3D patches centered on the lesion of interest (Barquero
et al., 2020; Maggi et al., 2020). In RimNet (Barquero et al., 2020), for
instance, an expert manually split conuent lesions, whereas Lou et al.
observed a consistent drop in performance in PRL classication in the
presence of conuent lesions (Lou et al., 2021). Although methods to
automatically split conuent lesions have been proposed (Dworkin
et al., 2019; Zhang et al., 2021), further developments are needed in
order to properly apply these in the presence of the CVS or PRL.
Finally, for every automated tool the regulatory environment re-
mains a critical barrier, as up to date less than 90 AI/ML-based medical
devices or algorithms have been approved by the US Food & Drugs
Administration (FDA). This challenge, however, is not unique to the
three biomarkers considered (Pinto et al., 2020) but shared also by
automated approaches segmenting WML or estimating brain atrophy.
4.3. Future research avenues
Standardization of the biomarkers’ assessment- The rst two
necessary steps toward the improvement of the above-referred ap-
proaches are the validation of the biomarkers’ specic criteria and
standardization of the relative MRI protocols. CL have been recently
included in the MS diagnostic criteria (Thompson et al., 2018), however,
a consensus on imaging and on their denition is still missing. In a
similar way, PRL urgently need a consensus denition and standardized
clinical protocols, whereas the initial criteria proposed for the CVS (Sati
et al., 2016) need to be updated in light of the latest studies. This would
clarify the automated methods’ goals, which so far have been extremely
dependent on specic expert labeling of each dataset or on the specic
criteria adopted.
Standardization and extensive validation of the automated
methods - Currently, it is difcult to compare the performance of
automated ML methods considering different criteria (such as the min-
imum lesion size) and being evaluated on private datasets. In the future,
the generalization of the proposed methods should be validated on large,
multi-site datasets with standardized metrics. For this purpose, we urge
research groups to organize grand challenges and release publicly
available datasets with manual annotations of CL, PRL, and CVS. As
already proved for several other tasks in medical imaging (Antonelli
et al., 2021), including for WML segmentation (Carass et al., 2017;
Commowick et al., 2018), such open data initiatives boost on the one
hand the development of state-of-the-art methods, and on the other
hand, help set benchmarks for a fair assessment. Only 5 of the 12
methods covered in this review are publicly available. In order to extend
their usage and foster a culture of open science, research groups should
make their code publicly available and possibly provide Docker (Docker,
2014)/Singularity (Kurtzer et al., 2017) images to facilitate their eval-
uation. Moreover, as successfully done for WML segmentation (Valverde
et al., 2019), domain-adaptation techniques should also be explored in
order to improve robustness of the automated ML methods to noise,
artifacts, and different protocols. So far, all three biomarkers have been
primarily studied at 3 T and 7 T, and therefore robust methods able to
work with images acquired at both magnetic eld strengths would be
very valuable. Machine learning algorithms could exploit 7 T enhanced
spatial resolution and tissue contrast by domain adaptation techniques
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
11
to improve their performance on 3 T imaging, which will continue to be
the main tool for clinicians as well as for clinical research and trials for
the foreseeable future. Although it would be highly desirable to have
methods that work also at the most accessible eld strength of 1.5 T, this
seems currently unlikely as the sensitivity to these biomarkers is eld-
dependent.
Transfer learning - Considering the scarcity of large, annotated
datasets, an additional strategy that should be explored consists of
transfer learning. Sharing of neural network weights between research
groups could foster interdisciplinary applicability of CNN trained on
relatively large datasets towards different purposes, such as CL, PRL, and
CVS, by ne-tuning the trained models in smaller datasets. Potential
advantages would include a shorter training time and robust feature
extraction across different MRI device manufacturers or different pulse
sequence acquisition parameters (Valverde et al., 2021).
Longitudinal assessment - Another possible research direction is an
expansion of the current methods to analyze longitudinal data. To the
best of our knowledge, only one study has tackled the automated lon-
gitudinal assessment of CL at 3 T (Fartaria et al., 2019), whereas PRL
evolution over time has not yet been assessed with automated ap-
proaches. CL are known to play a major role in disease progression
(Mainero et al., 2015) and considerable changes in their volume were
observed in longitudinal studies (Calabrese et al., 2008; Faizy et al.,
2019). Of similar interest, PRL and slowly-evolving lesions (SELs) vol-
ume assessment over time is a plausible future clinical measure of
treatment response (Absinta et al., 2021; Dal-Bianco et al., 2021; Elliott
et al., 2019; Elliott et al., 2019). Therefore, automated longitudinal
assessment of both CL and PRL could be of high relevance. Regarding
SELs, longitudinal WML segmentation approaches (Llad´
o et al., 2012)
could be adapted to track their evolution in a fully-automated way. This
would facilitate their assessment as currently, following an automated
cross-sectional WML segmentation, the lesion masks at each timepoint
are manually reviewed (Elliott et al., 2019).
Joint assessment of multiple biomarkers- To date, all the methods
proposed tackled the assessment of a single lesional biomarker, although
in the case of CL some methods consider WML as well (Fartaria et al.,
2019; La Rosa et al., 2020; Fartaria et al., 2016). Future work may aim at
automatically analyzing multiple biomarkers in a unied framework
(eg. with the same input images and algorithm) as this would be
extremely useful for research purposes or in clinics. Moreover, ML-based
algorithms have the potential to be useful also for prediction purposes. A
few automated methods based either on MRI (Tousignant et al., 2021;
Marzullo et al., 2019; Roca et al., 2020), optical coherence tomography
(Montolío et al., 2021), or clinical information (Pinto et al., 2020) have
already been presented to predict MS progression. Specically to the
biomarkers considered in the present review, Treaba et al. have pro-
posed a ML approach for the regression of both CL and PRL, in the same
cohort of patients, with disability progression (Treaba et al., 2021;3(3):
fcab134.). In this prospective, longitudinal study, the authors analyzed
brain scans of 100 MS patients using 7 T susceptibility-sensitive MRI in
which CL and PRL were segmented manually. Although the study had
some limitations, including the fact that the disability progression was
assessed only by the EDSS and only one ML-based method (gradient
boosting algorithm, XGBoost) was tested, it showed that 7 T MRI and the
combination of different biomarkers are promising in predicting MS
disability progression. Future studies should aim to combine the auto-
mated assessment of multiple biomarkers with clinical information and
other relevant markers to predict clinical outcomes or treatment effect.
Explainable AI - As discussed in this paper, ML methods combined
with specialized MRI sequences could play a fundamental role in sup-
porting the diagnosis of, and prognostication in, MS. However, the
complexity of DL algorithms hinders their interpretation, which has led
some to consider these methods as “black boxes.” The lack of an obvious
connection between biology, pathophysiology, and features revealed by
DL might diminish clinicians’ condence in these algorithms, again
hindering the adoption of such tools in clinical research and healthcare.
Explainable AI (XAI) methods are needed as to on one side provide
uncertainty estimates regarding the output provided and on the other
side transparency on the decisions taken by the DL-models. By
explainability, we refer to a set of domain features such as pixels of an
image or human-understandable high-level attributes that contribute to
the output decision of the model and its internal working. To our
knowledge, there are only two groups that have investigated XAI in MS.
Eitel et al. (Eitel et al., 2019) explored explainability to reveal relevant
voxel-wise locations that a trained CNN uses for distinguishing between
a normal and MS brain MRI. They found that diagnostic success relied on
the appearance of both lesions and non-lesional tissue (thalamus). Nair
et al.Nair et al. (2020) studied the uncertainty of DL-based lesion seg-
mentation to quantify the AI model reliability. Interestingly, their results
showed that discarding lesions with high estimated uncertainty from the
output segmentation would improve the performance of the model.
These two pioneering approaches strengthen the idea that explainability
and uncertainty measures can reliably provide new insights into how DL
models for MS work and potentially improve them and increase their
transparency.
Overall, we believe that developing explainable AI tools is crucial in
the ML MS research roadmap and would have an impact at both meth-
odological and clinical levels. First, explainable DL in MS would provide
new insights into model decisions and help identify any bias. Second, the
inclusion of uncertainty and explainability will help in increasing the
condence of clinicians considering their use, as well as improve the
quality of decision making and ultimately the clinical impact. Finally,
they may foster a better understanding of MS progression by generating
biologically interpretable measures of inammation and degeneration.
5. Conclusions
To summarize, automated or semi-automated ML-based approaches
aiming to segment and classify CL, CVS, and PRL are still in an early
stage. Nevertheless, these pioneering methods have the potential to
provide standardized identication of the biomarkers and facilitate their
large-scale assessment in clinical routines. Automated or semi-
automated tools could considerably reduce the current amount of time
and effort needed for a manual assessment. To date, however, some
limitations still hinder a broader adoption of these tools. First, there is a
general need for consensus criteria and standardized clinical protocols
for all three aforementioned biomarkers. Further, a major barrier to the
automated methods’ deployment is their lack of validation on multi-
center datasets acquired with different protocols. Future work should
focus on improving the robustness of the automated methods, extending
their framework with longitudinal data, and including interpretable
measures into their decisions. Finally, we encourage research groups to
organize grand challenges and release publicly available datasets. This
would boost the development of new methods and provide benchmarks
for a fair and standardized comparison that is currently lacking.
CRediT authorship contribution statement
Francesco La Rosa: Conceptualization, Methodology, Validation,
Formal analysis, Investigation, Data curation, Writing – original draft,
Writing – review & editing, Visualization. Maxence Wynen: Method-
ology, Visualization, Formal analysis, Investigation, Data curation,
Writing – original draft, Writing – review & editing. Omar Al-Louzi:
Methodology, Validation, Formal analysis, Data curation, Writing –
original draft, Writing – review & editing. Erin S Beck: Methodology,
Validation, Resources, Data curation, Writing – original draft, Writing –
review & editing, Visualization. Till Huelnhagen: Resources, Writing –
original draft, Writing – review & editing, Visualization. Pietro Maggi:
Resources, Writing – original draft, Writing – review & editing, Visual-
ization. Jean-Philippe Thiran: Investigation, Supervision, Project
administration, Funding acquisition. Tobias Kober: Investigation,
Writing – original draft, Writing – review & editing. Russell T
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
12
Shinohara: Validation, Investigation, Writing – original draft, Writing –
review & editing. Pascal Sati: Validation, Investigation, Writing –
original draft, Writing – review & editing. Daniel S Reich: Validation,
Formal analysis, Investigation, Resources, Writing – original draft,
Writing – review & editing. Cristina Granziera: Validation, Resources,
Writing – original draft. Martina Absinta: Validation, Formal analysis,
Investigation, Resources, Data curation, Writing – original draft, Writing
– review & editing, Visualization. Meritxell Bach Cuadra: Conceptu-
alization, Methodology, Validation, Investigation, Resources, Writing –
original draft, Writing – review & editing, Supervision, Project admin-
istration, Funding acquisition.
Declaration of Competing Interest
The University Hospital Basel (USB), as the employer of C.G., has
received the following fees which were used exclusively for research
support: (i) advisory board and consultancy fees from Actelion,
Genzyme-Sano, Novartis, GeNeuro and Roche; (ii) speaker feesfrom
Genzyme-Sano, Novartis, GeNeuro and Roche; (iii) research support
from Siemens, GeNeuro, Roche. M.A. has received consultancy fees from
GSK and Sano-Genzyme. P.M. has received support from Biogen and
Cliniques universitaires Saint-Luc Fonds de Recherche Clinique. D.S.R.
has received research support from Abata, Sano-Genzyme, and Vertex.
The other authors have no known competing nancial interests or per-
sonal relationships that could have appeared to inuence the work re-
ported in this paper.
Data availability
No data was used for the research described in the article.
Acknowledgements
F.L.R. is supported by the Swiss National Foundation (SNF) Postdoc
Mobility Fellowship (P500PB_206833), the European Union’s Horizon
2020 research and innovation program under the Marie Sklodowska-
Curie project TRABIT (agreement No 765148) and the Novartis Foun-
dation for Medical-Biological Research (#21A032). M.W. is supported
by a Swiss government excellence scholarship (#2021.0087). O.A. is
supported by a National Multiple Sclerosis Society (NMSS) - American
Brain Foundation Clinician Scientist Development Award (FAN-
1807–32163). E.S.B. is supported by a Career Transition Fellowship
from the National Multiple Sclerosis Society. P.S., O.A., E.S.B., and D.S.
R. are supported by the Intramural Research Program of the National
Institute of Neurological Disorders and Stroke, National Institutes of
Health, Bethesda, Maryland, USA. R.T.S. is partially supported by
R01NS060910, R01MH112847, R01MH123550, U01NS116776, and
R01NS112274 from the National Institutes of Health. The content is
solely the responsibility of the authors and does not necessarily repre-
sent the ofcial views of the National Institutes of Health. M.A. is sup-
ported by the Conrad N. Hilton Foundation (Marilyn Hilton Bridging
Award for Physician-Scientists, grant #17313), the International Pro-
gressive MS alliance, the Roche Foundation for Independent Research,
the Cariplo Foundation (grant #1677), and the FRRB Early Career
Award (grant#1750327).
We acknowledge access to the facilities and expertise of the CIBM
Center for Biomedical Imaging, a Swiss research center of excellence
founded and supported by Lausanne University Hospital (CHUV), Uni-
versity of Lausanne (UNIL), Ecole Polytechnique fed´
erale de Lausanne
(EPFL), University of Geneva (UNIGE) and Geneva University Hospitals
(HUG). We thank Thomas Yu for proofreading the manuscript.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.nicl.2022.103205.
References
Absinta, M., Sati, P., Schindler, M., et al., 2016. Persistent 7-tesla phase rim predicts poor
outcome in new multiple sclerosis patient lesions. J Clin Invest. 126 (7), 2597–2609.
https://doi.org/10.1172/JCI86198.
Absinta, M., Sati, P., Fechner, A., Schindler, M.K., Nair, G., Reich, D.S., 2018.
Identication of chronic active multiple sclerosis lesions on 3T MRI. Am J
Neuroradiol. 39 (7), 1233–1238. https://doi.org/10.3174/ajnr.A5660.
Absinta, M., Sati, P., Masuzzo, F., et al., 2019. Association of chronic active multiple
sclerosis lesions with disability in vivo. JAMA Neurol. 76 (12), 1474. https://doi.
org/10.1001/jamaneurol.2019.2399.
Absinta, M., Maric, D., Gharagozloo, M., et al., 2021. A lymphocyte–microglia–astrocyte
axis in chronic active multiple sclerosis. Nature 597 (7878), 709–714. https://doi.
org/10.1038/s41586-021-03892-7.
Antonelli M, Reinke A, Bakas S, et al. The Medical Segmentation Decathlon.
ArXiv210605735 Cs Eess. Published online June 10, 2021. Accessed November 8,
2021. http://arxiv.org/abs/2106.05735.
Barkhof, F., 2002. The clinico-radiological paradox in multiple sclerosis revisited. Curr.
Opin. Neurol. 15 (3), 239–245. https://doi.org/10.1097/00019052-200206000-
00003.
Barquero, G., La Rosa, F., Kebiri, H., et al., 2020. RimNet: A deep 3D multimodal MRI
architecture for paramagnetic rim lesion assessment in multiple sclerosis.
NeuroImage Clin. 28, 102412 https://doi.org/10.1016/j.nicl.2020.102412.
Beck ES, Maranzano J, Luciano NJ, et al. Cortical lesion hotspots and association of
subpial lesions with disability in multiple sclerosis. Mult Scler J. Published online
February 10, 2022:135245852110691. doi:10.1177/13524585211069167.
Beck, E.S., Gai, N., Filippini, S., Maranzano, J., Nair, G., Reich, D.S., 2020. Inversion
recovery susceptibility weighted imaging with enhanced T2 Weighting at 3 T
improves visualization of subpial cortical multiple sclerosis lesions. Invest. Radiol.
55 (11), 727–735. https://doi.org/10.1097/RLI.0000000000000698.
Calabrese, M., Filippi, M., Rovaris, M., et al., 2008. Morphology and evolution of cortical
lesions in multiple sclerosis. A longitudinal MRI study. NeuroImage. 42 (4),
1324–1328. https://doi.org/10.1016/j.neuroimage.2008.06.028.
Calabrese, M., Filippi, M., Gallo, P., 2010. Cortical lesions in multiple sclerosis. Nat Rev
Neurol. 6 (8), 438–444. https://doi.org/10.1038/nrneurol.2010.93.
Calabrese, M., Poretto, V., Favaretto, A., et al., 2012. Cortical lesion load associates with
progression of disability in multiple sclerosis. Brain. 135 (10), 2952–2961. https://
doi.org/10.1093/brain/aws246.
Calabrese, M., Romualdi, C., Poretto, V., et al., 2013. The changing clinical course of
multiple sclerosis: A matter of gray matter: Cortical Pathology and Secondary
Progressive MS. Ann. Neurol. 74 (1), 76–83. https://doi.org/10.1002/ana.23882.
Campion, T., Smith, R.J.P., Altmann, D.R., et al., 2017. FLAIR* to visualize veins in white
matter lesions: A new tool for the diagnosis of multiple sclerosis? Eur. Radiol. 27
(10), 4257–4263. https://doi.org/10.1007/s00330-017-4822-z.
Carass, A., Roy, S., Jog, A., et al., 2017. Longitudinal multiple sclerosis lesion
segmentation: Resource and challenge. NeuroImage. 148, 77–102. https://doi.org/
10.1016/j.neuroimage.2016.12.064.
Castellaro, M., Tamanti, A., Pisani, A.I., Pizzini, F.B., Crescenzo, F., Calabrese, M., 2020.
The Use of the Central Vein Sign in the Diagnosis of Multiple Sclerosis: A Systematic
Review and Meta-analysis. Diagnostics. 10 (12), 1025. https://doi.org/10.3390/
diagnostics10121025.
Ciotti, J.R., Eby, N.S., Brier, M.R., et al., 2022. Central vein sign and other radiographic
features distinguishing myelin oligodendrocyte glycoprotein antibody disease from
multiple sclerosis and aquaporin-4 antibody-positive neuromyelitis optica. Mult
Scler J. 28 (1), 49–60. https://doi.org/10.1177/13524585211007086.
Clarke, M.A., Pareto, D., Pessini-Ferreira, L., et al., 2020. Value of 3T Susceptibility-
Weighted Imaging in the Diagnosis of Multiple Sclerosis. Am J Neuroradiol. 41 (6),
1001–1008. https://doi.org/10.3174/ajnr.A6547.
Commowick O, Cervenansky F, Cotton F, Dojat M. MSSEG-2 challenge proceedings:
Multiple sclerosis new lesions segmentation challenge using a data management and
processing infrastructure. In: ; 2021:1. Accessed October 28, 2021. https://hal.inria.
fr/hal-03358968.
Commowick, O., Istace, A., Kain, M., et al., 2018. Objective evaluation of multiple
sclerosis lesion segmentation using a data management and processing
infrastructure. Sci. Rep. 8 (1), 13650. https://doi.org/10.1038/s41598-018-31911-
7.
Cortese, R., Magnollay, L., Tur, C., et al., 2018. Value of the central vein sign at 3T to
differentiate MS from seropositive NMOSD. Neurology. 90 (14), e1183–e1190.
https://doi.org/10.1212/WNL.0000000000005256.
Cortese, R., Collorone, S., Ciccarelli, O., Toosy, A.T., 2019. Advances in brain imaging in
multiple sclerosis. Ther Adv Neurol Disord. 12 https://doi.org/10.1177/
1756286419859722.
Dal-Bianco, A., Grabner, G., Kronnerwetter, C., et al., 2021. Long-term evolution of
multiple sclerosis iron rim lesions in 7 T MRI. Brain. 144 (3), 833–847. https://doi.
org/10.1093/brain/awaa436.
Merkel D. Docker: lightweight Linux containers for consistent development and
deployment. Linux J. Published online March 1, 2014. Accessed March 15, 2022.
https://dl.acm.org/doi/abs/10.5555/2600239.2600241.
Dworkin, J.D., Sati, P., Solomon, A., et al., 2018. Automated integration of multimodal
MRI for the probabilistic detection of the central vein sign in white matter lesions.
Am J Neuroradiol. 39 (10), 1806–1813. https://doi.org/10.3174/ajnr.A5765.
Dworkin, J.D., Linn, K.A., Oguz, I., et al., 2019. An automated statistical technique for
counting distinct multiple sclerosis lesions. Am J Neuroradiol. 39 (4), 626–633.
https://doi.org/10.3174/ajnr.A5556.
Eitel, F., Soehler, E., Bellmann-Strobl, J., et al., 2019. Uncovering convolutional neural
network decisions for diagnosing multiple sclerosis on conventional MRI using layer-
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
13
wise relevance propagation. NeuroImage Clin. 24, 102003 https://doi.org/10.1016/
j.nicl.2019.102003.
Elliott, C., Wolinsky, J.S., Hauser, S.L., et al., 2019. Slowly expanding/evolving lesions as
a magnetic resonance imaging marker of chronic active multiple sclerosis lesions.
Mult Scler J. 25 (14), 1915–1925. https://doi.org/10.1177/1352458518814117.
Elliott, C., Belachew, S., Wolinsky, J.S., et al., 2019. Chronic white matter lesion activity
predicts clinical progression in primary progressive multiple sclerosis. Brain. 142
(9), 2787–2799. https://doi.org/10.1093/brain/awz212.
Elliott, C., Belachew, S., Fisher, E., et al., 2021. MRI Characteristics of Chronic MS
Lesions by Phase Rim Detection and/or Slowly Expanding Properties. Neurology. 96
(15 Supplement), 4101.
Faizy, T.D., Thaler, C., Ceyrowski, T., et al., 2017. Reliability of cortical lesion detection
on double inversion recovery MRI applying the MAGNIMS-Criteria in multiple
sclerosis patients within a 16-months period. Kassubek J, ed. PLOS ONE. https://doi.
org/10.1371/journal.pone.0172923.
Faizy, T.D., Broocks, G., Thaler, C., et al., 2019. Development of Cortical Lesion Volumes
on Double Inversion Recovery MRI in Patients With Relapse-Onset Multiple
Sclerosis. Front. Neurol. 10 https://doi.org/10.3389/fneur.2019.00133.
Fartaria, M.J., Bonnier, G., Roche, A., et al., 2016. Automated detection of white matter
and cortical lesions in early stages of multiple sclerosis: Automated MS Lesion
Segmentation. J. Magn. Reson. Imaging 43 (6), 1445–1454. https://doi.org/
10.1002/jmri.25095.
Fartaria MJ, Roche A, Meuli R, Granziera C, Kober T, Bach Cuadra M. Segmentation of
Cortical and Subcortical Multiple Sclerosis Lesions Based on Constrained Partial
Volume Modeling. In: Medical Image Computing and Computer Assisted Intervention −
MICCAI 2017. Vol 10435. Lecture Notes in Computer Science. Springer International
Publishing; 2017:142-149. doi:10.1007/978-3-319-66179-7_17.
Fartaria, M.J., Kober, T., Granziera, C., Bach, C.M., 2019. Longitudinal analysis of white
matter and cortical lesions in multiple sclerosis. NeuroImage Clin. 23, 101938
https://doi.org/10.1016/j.nicl.2019.101938.
Fartaria, M.J., Sati, P., Todea, A., et al., 2019. Automated Detection and Segmentation of
Multiple Sclerosis Lesions Using Ultra–High-Field MP2RAGE. Invest. Radiol. 54 (6),
356–364. https://doi.org/10.1097/RLI.0000000000000551.
Favaretto, A., Lazzarotto, A., Poggiali, D., et al., 2016. MRI-detectable cortical lesions in
the cerebellum and their clinical relevance in multiple sclerosis. Mult Scler J. 22 (4),
494–501. https://doi.org/10.1177/1352458515594043.
Filippi, M., Preziosa, P., Banwell, B.L., et al., 2019. Assessment of lesions on magnetic
resonance imaging in multiple sclerosis: practical guidelines. Brain. 142 (7),
1858–1875. https://doi.org/10.1093/brain/awz144.
Frangi AF, Niessen WJ, Vincken KL, Viergever MA. Multiscale vessel enhancement
ltering. In: Medical Image Computing and Computer-Assisted Intervention —
MICCAI’98. Vol 1496. Lecture Notes in Computer Science. Springer Berlin
Heidelberg; 1998:130-137. doi:10.1007/BFb0056195.
Frischer, J.M., Weigand, S.D., Guo, Y., et al., 2015. Clinical and pathological insights into
the dynamic nature of the white matter multiple sclerosis plaque: dynamic nature of
MS Plaque. Ann. Neurol. 78 (5), 710–721. https://doi.org/10.1002/ana.24497.
Gait´
an, M.I., Ya˜
nez, P., Paday Formenti, M.E., et al., 2020. SWAN-Venule: An optimized
MRI technique to detect the central vein sign in MS Plaques. Am J Neuroradiol. 41
(3), 456–460. https://doi.org/10.3174/ajnr.A6437.
Geurts JJG, Bo L, Pouwels PJW, Castelijns JA, Polman CH, Barkhof F. Cortical Lesions in
Multiple Sclerosis: Combined Postmortem MR Imaging and Histopathology.
Published online 2005:6.
Geurts, J.J.G., Roosendaal, S.D., Calabrese, M., et al., 2011. Consensus recommendations
for MS cortical lesion scoring using double inversion recovery MRI. Neurology. 76
(5), 418–424. https://doi.org/10.1212/WNL.0b013e31820a0cc4.
Giorgio, A., Stromillo, M.L., Rossi, F., et al., 2011. Cortical lesions in radiologically
isolated syndrome. Neurology. 77 (21), 1896–1899. https://doi.org/10.1212/
WNL.0b013e318238ee9b.
Giorgio, A., Stromillo, M.L., Bartolozzi, M.L., et al., 2014. Relevance of hypointense brain
MRI lesions for long-term worsening of clinical disability in relapsing multiple
sclerosis. Mult Scler J. 20 (2), 214–219. https://doi.org/10.1177/
1352458513494490.
Granziera, C., Wuerfel, J., Barkhof, F., et al., 2021. Quantitative magnetic resonance
imaging towards clinical application in multiple sclerosis. Brain. 144 (5),
1296–1311. https://doi.org/10.1093/brain/awab029.
Haacke, E.M., Mittal, S., Wu, Z., Neelavalli, J., Cheng, Y.C.N., 2009. Susceptibility-
weighted imaging: Technical aspects and clinical applications, Part 1. Am J
Neuroradiol. 30 (1), 19–30. https://doi.org/10.3174/ajnr.A1400.
Hagens, M.H., Burggraaff, J., Kilsdonk, I.D., et al., 2019. Impact of 3 Tesla MRI on
interobserver agreement in clinically isolated syndrome: A MAGNIMS multicentre
study. Mult Scler J. 25 (3), 352–360. https://doi.org/10.1177/1352458517751647.
Harrison, D.M., Roy, S., Oh, J., et al., 2015. Association of cortical lesion burden on 7-T
Magnetic Resonance imaging with cognition and disability in multiple sclerosis.
JAMA Neurol. 72 (9), 1004. https://doi.org/10.1001/jamaneurol.2015.1241.
Hemond, C.C., Bakshi, R., 2018. Magnetic resonance imaging in multiple sclerosis. Cold
Spring Harb Perspect Med. 8 (5), a028969 https://doi.org/10.1101/cshperspect.
a028969.
Huang W, Sweeney EM, Kaunzner UW, Wang Y, Gauthier SA, Nguyen TD. Quantitative
susceptibility mapping versus phase imaging to identify multiple sclerosis iron rim
lesions with demyelination. J Neuroimaging. Published online March 9, 2022:
jon.12987. doi:10.1111/jon.12987.
Ineichen, B.V., Beck, E.S., Piccirelli, M., Reich, D.S., 2021. New Prospects for ultra-high-
eld magnetic resonance imaging in multiple sclerosis. Invest. Radiol. 56 (11),
773–784. https://doi.org/10.1097/RLI.0000000000000804.
Junker, A., Wozniak, J., Voigt, D., et al., 2020. Extensive subpial cortical demyelination
is specic to multiple sclerosis. Brain Pathol. 30 (3), 641–652. https://doi.org/
10.1111/bpa.12813.
Kaisey, M., Solomon, A.J., Luu, M., Giesser, B.S., Sicotte, N.L., 2019. Incidence of
multiple sclerosis misdiagnosis in referrals to two academic centers. Mult Scler Relat
Disord. 30, 51–56. https://doi.org/10.1016/j.msard.2019.01.048.
Kau, T., Taschwer, M., Deutschmann, H., Sch¨
onfelder, M., Weber, J.R., Hausegger, K.A.,
2013. The “central vein sign”: is there a place for susceptibility weighted imaging in
possible multiple sclerosis? Eur. Radiol. 23 (7), 1956–1962. https://doi.org/
10.1007/s00330-013-2791-4.
Kaur, A., Kaur, L., Singh, A., 2021. State-of-the-Art Segmentation Techniques and Future
Directions for Multiple Sclerosis Brain Lesions. Arch. Comput. Methods Eng. 28 (3),
951–977. https://doi.org/10.1007/s11831-020-09403-7.
Kidd, D., Barkhof, F., McConnell, R., Algra, P.R., Allen, I.V., Revesz, T., 1999. Cortical
lesions in multiple sclerosis. Brain. 122 (1), 17–26. https://doi.org/10.1093/brain/
122.1.17.
Kilsdonk, I.D., Jonkman, L.E., Klaver, R., et al., 2016. Increased cortical grey matter
lesion detection in multiple sclerosis with 7 T MRI: a post-mortem verication study.
Brain. 139 (5), 1472–1481. https://doi.org/10.1093/brain/aww037.
Kolb, H., Absinta, M., Beck, E.S., et al., 2021. 7T MRI Differentiates Remyelinated from
Demyelinated Multiple Sclerosis Lesions. Ann. Neurol. 90 (4), 612–626. https://doi.
org/10.1002/ana.26194.
Kurtzer, G.M., Sochat, V., Singularity, B.MW., 2017. Scientic containers for mobility of
compute. Gursoy A, ed. PLOS ONE. https://doi.org/10.1371/journal.pone.0177459.
La Rosa F, Fartaria MJ, Kober T, et al. Shallow vs Deep Learning Architectures for White
Matter Lesion Segmentation in the Early Stages of Multiple Sclerosis. In: Crimi A,
Bakas S, Kuijf H, Keyvan F, Reyes M, van Walsum T, eds. Brainlesion: Glioma, Multiple
Sclerosis, Stroke and Traumatic Brain Injuries. Vol 11383. Lecture Notes in Computer
Science. Springer International Publishing; 2019:142-151. doi:10.1007/978-3-030-
11723-8_14.
La Rosa F, Beck ES, Abdulkadir A, et al. Automated Detection of Cortical Lesions in
Multiple Sclerosis Patients with 7T MRI. In: Medical Image Computing and Computer
Assisted Intervention – MICCAI 2020. Vol 12264. Lecture Notes in Computer Science.
Springer International Publishing; 2020:584-593. doi:10.1007/978-3-030-59719-1_
57.
La Rosa F, Abdulkadir A, Fartaria MJ, et al. Multiple sclerosis cortical and WM lesion
segmentation at 3T MRI: a deep learning method based on FLAIR and MP2RAGE.
NeuroImage Clin. Published online June 2020:102335. doi:10.1016/j.
nicl.2020.102335.
La Rosa F, Beck ES, Maranzano J, et al. Multiple sclerosis cortical lesion detection with
deep learning at ultra-high-eld MRI. NMR Biomed. Published online March 31,
2022. doi:10.1002/nbm.4730.
Llad´
o, X., Oliver, A., Cabezas, M., et al., 2012. Segmentation of multiple sclerosis lesions
in brain MRI: A review of automated approaches. Inf. Sci. 186 (1), 164–185. https://
doi.org/10.1016/j.ins.2011.10.011.
Lou C, Sati P, Absinta M, et al. Fully Automated Detection of Paramagnetic Rims in
Multiple Sclerosis Lesions on 3T Susceptibility-Based MR Imaging. NeuroImage Clin.
Published online August 2021:102796. doi:10.1016/j.nicl.2021.102796.
Lu, P.J., Yoo, Y., Rahmanzadeh, R., et al., 2021. GAMER MRI: Gated-attention
mechanism ranking of multi-contrast MRI in brain pathology. NeuroImage Clin. 29,
102522 https://doi.org/10.1016/j.nicl.2020.102522.
Luchetti, S., Fransen, N.L., van Eden, C.G., Ramaglia, V., Mason, M., Huitinga, I., 2018.
Progressive multiple sclerosis patients show substantial lesion activity that correlates
with clinical disease severity and sex: a retrospective autopsy cohort analysis. Acta
Neuropathol (Berl). 135 (4), 511–528. https://doi.org/10.1007/s00401-018-1818-y.
Madsen, M.A.J., Wiggermann, V., Bramow, S., Christensen, J.R., Sellebjerg, F.,
Siebner, H.R., 2021. Imaging Cortical Multiple Sclerosis Lesions with Ultra-High
Field MRI. Radiology and Imaging. https://doi.org/10.1101/2021.06.25.21259363.
Maggi, P., Absinta, M., Grammatico, M., et al., 2018. Central vein sign differentiates
Multiple Sclerosis from central nervous system inammatory vasculopathies: Central
Vein Sign. Ann. Neurol. 83 (2), 283–294. https://doi.org/10.1002/ana.25146.
Maggi, P., Fartaria, M.J., Jorge, J., et al., 2020. CVSnet: A machine learning approach for
automated central vein sign assessment in multiple sclerosis. NMR Biomed. 33 (5)
https://doi.org/10.1002/nbm.4283.
Maggi, P., Sati, P., Nair, G., et al., 2020. Paramagnetic Rim Lesions are Specic to
Multiple Sclerosis: An International Multicenter 3T MRI Study. Ann. Neurol. 88 (5),
1034–1042. https://doi.org/10.1002/ana.25877.
Maggi, P., Kuhle, J., Sch¨
adelin, S., et al., 2021. Chronic White Matter Inammation and
Serum Neurolament Levels in Multiple Sclerosis. Neurology. 97 (6), e543–e553.
https://doi.org/10.1212/WNL.0000000000012326.
Mainero, C., Louapre, C., Govindarajan, S.T., et al., 2015. A gradient in cortical
pathology in multiple sclerosis by in vivo quantitative 7 T imaging. Brain. 138 (4),
932–945. https://doi.org/10.1093/brain/awv011.
Maranzano, J., Dadar, M., Rudko, D.A., et al., 2019. Comparison of Multiple Sclerosis
Cortical Lesion Types Detected by Multicontrast 3T and 7T MRI. Am J Neuroradiol.
40 (7), 1162–1169. https://doi.org/10.3174/ajnr.A6099.
Martínez-Heras, E., Solana, E., Prados, F., et al., 2020. Characterization of multiple
sclerosis lesions with distinct clinical correlates through quantitative diffusion MRI.
NeuroImage Clin. 28, 102411 https://doi.org/10.1016/j.nicl.2020.102411.
Marzullo, A., Kocevar, G., Stamile, C., 2019. Prediction of multiple sclerosis patient
disability from structural connectivity using convolutional neural networks. In, et al.
41st Annual International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBC). IEEE 2019, 2087–2090. https://doi.org/10.1109/
EMBC.2019.8856845.
Mistry, N., Abdel-Fahim, R., Samaraweera, A., et al., 2016. Imaging central veins in brain
lesions with 3-T T2*-weighted magnetic resonance imaging differentiates multiple
F. La Rosa et al.
NeuroImage: Clinical 36 (2022) 103205
14
sclerosis from microangiopathic brain lesions. Mult Scler J. 22 (10), 1289–1296.
https://doi.org/10.1177/1352458515616700.
Mittal, S., Wu, Z., Neelavalli, J., Haacke, E.M., 2009. Susceptibility-Weighted Imaging:
Technical Aspects and Clinical Applications, Part 2. Am J Neuroradiol. 30 (2),
232–252. https://doi.org/10.3174/ajnr.A1461.
Montolío, A., Martín-Gallego, A., Cego˜
nino, J., et al., 2021. Machine learning in
diagnosis and disability prediction of multiple sclerosis using optical coherence
tomography. Comput. Biol. Med. 133, 104416 https://doi.org/10.1016/j.
compbiomed.2021.104416.
Müller, J., La Rosa, F., Beaumont, J., et al., 2022. FLAWS: New Sensitive 3T MRI
Contrasts for Cortical Lesion Detection in Multiple Sclerosis. Published online, Invest
Radiol.
Nair, T., Precup, D., Arnold, D.L., Arbel, T., 2020. Exploring uncertainty measures in
deep networks for Multiple sclerosis lesion detection and segmentation. Med. Image
Anal. 59, 101557 https://doi.org/10.1016/j.media.2019.101557.
Nielsen, A.S., Kinkel, R.P., Tinelli, E., Benner, T., Cohen-Adad, J., Mainero, C., 2012.
Focal cortical lesion detection in multiple sclerosis: 3 tesla DIR versus 7 tesla FLASH-
T2*. J. Magn. Reson. Imaging 35 (3), 537–542. https://doi.org/10.1002/
jmri.22847.
Nielsen, A.S., Kinkel, R.P., Madigan, N., Tinelli, E., Benner, T., Mainero, C., 2013.
Contribution of cortical lesion subtypes at 7T MRI to physical and cognitive
performance in MS. Neurology. 81 (7), 641–649. https://doi.org/10.1212/
WNL.0b013e3182a08ce8.
Oladosu, O., Liu, W.Q., Pike, B.G., Koch, M., Metz, L.M., Zhang, Y., 2021. Advanced
Analysis of Diffusion Tensor Imaging Along With Machine Learning Provides New
Sensitive Measures of Tissue Pathology and Intra-Lesion Activity in Multiple
Sclerosis. Front. Neurosci. 15, 634063 https://doi.org/10.3389/fnins.2021.634063.
Ontaneda, D., Sati, P., Raza, P., et al., 2021. Central vein sign: A diagnostic biomarker in
multiple sclerosis (CAVS-MS) study protocol for a prospective multicenter trial.
NeuroImage Clin. 32, 102834 https://doi.org/10.1016/j.nicl.2021.102834.
Pinto, M.F., Oliveira, H., Batista, S., et al., 2020. Prediction of disease progression and
outcomes in multiple sclerosis with machine learning. Sci. Rep. 10 (1), 21038.
https://doi.org/10.1038/s41598-020-78212-6.
Preziosa, P., Pagani, E., Meani, A., et al., 2022. Slowly Expanding Lesions Predict 9-Year
Multiple Sclerosis Disease Progression. Neurol - Neuroimmunol Neuroinammation.
9 (2), e1139 https://doi.org/10.1212/NXI.0000000000001139.
Reich DS, Lucchinetti CF, Calabresi PA. Multiple Sclerosis. Longo DL, ed. N Engl J Med.
2018;378(2):169-180. doi:10.1056/NEJMra1401483.
Roca, P., Attye, A., Colas, L., et al., 2020. Articial intelligence to predict clinical
disability in patients with multiple sclerosis using FLAIR MRI. Diagn Interv Imaging.
101 (12), 795–802. https://doi.org/10.1016/j.diii.2020.05.009.
Samaraweera, A.P.R., Clarke, M.A., Whitehead, A., et al., 2017. The Central Vein Sign in
Multiple Sclerosis Lesions Is Present Irrespective of the T2* Sequence at 3 T: The
Central Vein Sign in MS. J. Neuroimaging 27 (1), 114–121. https://doi.org/
10.1111/jon.12367.
Sati, P., George, I.C., Shea, C.D., Gait´
an, M.I., Reich, D.S., 2012. FLAIR*: A Combined MR
Contrast Technique for Visualizing White Matter Lesions and Parenchymal Veins.
Radiology 265 (3), 926–932. https://doi.org/10.1148/radiol.12120208.
Sati, P., Thomasson, D., Li, N., et al., 2014. Rapid, high-resolution, whole-brain,
susceptibility-based MRI of multiple sclerosis. Mult Scler J. 20 (11), 1464–1470.
https://doi.org/10.1177/1352458514525868.
Sati, P., Oh, J., Constable, R.T., et al., 2016. The central vein sign and its clinical
evaluation for the diagnosis of multiple sclerosis: a consensus statement from the
North American Imaging in Multiple Sclerosis Cooperative. Nat Rev Neurol. 12 (12),
714–722. https://doi.org/10.1038/nrneurol.2016.166.
Scalfari, A., Romualdi, C., Nicholas, R.S., et al., 2018. The cortical damage, early
relapses, and onset of the progressive phase in multiple sclerosis. Neurology. 90 (24),
e2107–e2118. https://doi.org/10.1212/WNL.0000000000005685.
Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-Scale Image
Recognition. ArXiv14091556 Cs. Published online April 10, 2015. Accessed May 7,
2020. http://arxiv.org/abs/1409.1556.
Sinnecker, T., Dorr, J., Pfueller, C.F., et al., 2012. Distinct lesion morphology at 7-T MRI
differentiates neuromyelitis optica from multiple sclerosis. Neurology. 79 (7),
708–714. https://doi.org/10.1212/WNL.0b013e3182648bc8.
Sinnecker, T., Clarke, M.A., Meier, D., et al., 2019. Evaluation of the Central Vein Sign as
a Diagnostic Imaging Biomarker in Multiple Sclerosis. JAMA Neurol. 76 (12), 1446.
https://doi.org/10.1001/jamaneurol.2019.2478.
Solomon, A.J., Schindler, M.K., Howard, D.B., et al., 2016. “Central vessel sign” on 3T
FLAIR* for the differentiation of multiple sclerosis from migraine. Ann. Clin. Transl.
Neurol. 3 (2), 82–87. https://doi.org/10.1002/acn3.273.
Solomon, A.J., Bourdette, D.N., Cross, A.H., et al., 2016. The contemporary spectrum of
multiple sclerosis misdiagnosis: A multicenter study. Neurology. 87 (13),
1393–1399. https://doi.org/10.1212/WNL.0000000000003152.
Solomon, A.J., Watts, R., Ontaneda, D., Absinta, M., Sati, P., Reich, D.S., 2018.
Diagnostic performance of central vein sign for multiple sclerosis with a simplied
three-lesion algorithm. Mult Scler J. 24 (6), 750–757. https://doi.org/10.1177/
1352458517726383.
Solomon, A.J., Naismith, R.T., Cross, A.H., 2019. Misdiagnosis of multiple sclerosis:
Impact of the 2017 McDonald criteria on clinical practice. Neurology. 92 (1), 26–33.
https://doi.org/10.1212/WNL.0000000000006583.
Sparacia, G., Agnello, F., Gambino, A., Sciortino, M., Midiri, M., 2018. Multiple sclerosis:
High prevalence of the ‘central vein’ sign in white matter lesions on susceptibility-
weighted images. Neuroradiol J. 31 (4), 356–361. https://doi.org/10.1177/
1971400918763577.
Suh, C.H., Kim, S.J., Jung, S.C., Choi, C.G., Kim, H.S., 2019. The “Central Vein Sign” on
T2*-weighted Images as a Diagnostic Tool in Multiple Sclerosis: A Systematic Review
and Meta-analysis using Individual Patient Data. Sci. Rep. 9 (1), 18188. https://doi.
org/10.1038/s41598-019-54583-3.
Tallantyre, E.C., Dixon, J.E., Donaldson, I., et al., 2011. Ultra-high-eld imaging
distinguishes MS lesions from asymptomatic white matter lesions. Neurology. 76 (6),
534–539. https://doi.org/10.1212/WNL.0b013e31820b7630.
Tardif, C.L., Bedell, B.J., Eskildsen, S.F., Collins, D.L., Pike, G.B., 2012. Quantitative
Magnetic Resonance Imaging of Cortical Multiple Sclerosis Pathology. Mult Scler Int.
2012, 1–13. https://doi.org/10.1155/2012/742018.
Tardif CL, Collins DL, Eskildsen SF, Richardson JB, Pike GB. Segmentation of Cortical MS
Lesions on MRI Using Automated Laminar Prole Shape Analysis. In: Medical Image
Computing and Computer-Assisted Intervention – MICCAI 2010. Vol 6363. Lecture
Notes in Computer Science. Springer Berlin Heidelberg; 2010:181-188. doi:10.1007/
978-3-642-15711-0_23.
Thaler, C., Kyselyova, A.A., Faizy, T.D., et al., 2021. Heterogeneity of multiple sclerosis
lesions in fast diffusional kurtosis imaging. Bergsland N, ed. PLOS ONE. https://doi.
org/10.1371/journal.pone.0245844.
Thompson, A.J., Banwell, B.L., Barkhof, F., et al., 2018. Diagnosis of multiple sclerosis:
2017 revisions of the McDonald criteria. Lancet Neurol. 17 (2), 162–173. https://
doi.org/10.1016/S1474-4422(17)30470-2.
Tousignant A, Lemaître P, Precup D, Arnold DL, Arbel T. Prediction of Disease
Progression in Multiple Sclerosis Patients using Deep Learning Analysis of MRI Data.
Accessed November 24, 2021. https://proceedings.mlr.press/v102/tousignant19a.
html.
Tranfa, M., Tortora, M., Pontillo, G., et al., 2022. The central vein sign helps in
differentiating multiple sclerosis from its mimickers: lessons from Fabry disease. Eur.
Radiol. 32 (6), 3846–3854. https://doi.org/10.1007/s00330-021-08487-4.
Treaba, C.A., Granberg, T.E., Sormani, M.P., et al., 2019. Longitudinal Characterization
of Cortical Lesion Development and Evolution in Multiple Sclerosis with 7.0-T MRI.
Radiology 291 (3), 740–749. https://doi.org/10.1148/radiol.2019181719.
Treaba, C.A., Conti, A., Klawiter, E.C., et al., 2021;3(3):fcab134.. Cortical and phase rim
lesions on 7 T MRI as markers of multiple sclerosis disease progression. Brain
Commun. https://doi.org/10.1093/braincomms/fcab134.
Valcarcel, A.M., Linn, K.A., Khalid, F., et al., 2018. A dual modeling approach to
automatic segmentation of cerebral T2 hyperintensities and T1 black holes in
multiple sclerosis. NeuroImage Clin. 20, 1211–1221. https://doi.org/10.1016/j.
nicl.2018.10.013.
Valcarcel, A.M., Linn, K.A., Vandekar, S.N., et al., 2018. MIMoSA: An Automated Method
for Intermodal Segmentation Analysis of Multiple Sclerosis Brain Lesions: Method
For Inter-Modal Segmentation Analysis. J. Neuroimaging 28 (4), 389–398. https://
doi.org/10.1111/jon.12506.
Valverde, J.M., Imani, V., Abdollahzadeh, A., et al., 2021. Transfer Learning in Magnetic
Resonance Brain Imaging: A Systematic Review. J Imaging. 7 (4), 66. https://doi.
org/10.3390/jimaging7040066.
Valverde, S., Salem, M., Cabezas, M., et al., 2019. One-shot domain adaptation in
multiple sclerosis lesion segmentation using convolutional neural networks.
NeuroImage Clin. 21, 101638 https://doi.org/10.1016/j.nicl.2018.101638.
Vrenken H, Jenkinson M, Pham D, et al. Opportunities for Understanding MS
Mechanisms and Progression With MRI Using Large-Scale Data Sharing and Articial
Intelligence. Neurology. Published online October 4, 2021:10.1212/
WNL.0000000000012884. doi:10.1212/WNL.0000000000012884.
Wattjes, M.P., Ciccarelli, O., Reich, D.S., et al., 2021. 2021 MAGNIMS–CMSC–NAIMS
consensus recommendations on the use of MRI in patients with multiple sclerosis.
Lancet Neurol. 20 (8), 653–670. https://doi.org/10.1016/S1474-4422(21)00095-8.
Ye, Z., George, A., Wu, A.T., et al., 2020. Deep learning with diffusion basis spectrum
imaging for classication of multiple sclerosis lesions. Ann. Clin. Transl. Neurol. 7
(5), 695–706. https://doi.org/10.1002/acn3.51037.
Zeng, C., Gu, L., Liu, Z., Zhao, S., 2020. Review of Deep Learning Approaches for the
Segmentation of Multiple Sclerosis Lesions on Brain MRI. Front Neuroinformatics.
14, 610967 https://doi.org/10.3389/fninf.2020.610967.
Zhang H, Zhang J, Yang G, Spincemaille P, Nguyen TD, Wang Y. Memory U-Net:
Memorizing Where to Vote for Lesion Instance Segmentation. In: ; 2021. Accessed
November 19, 2021. https://openreview.net/forum?id=JbWMYLN5Hba.
Zhang, H., Nguyen, T.D., Zhang, J., et al., 2022. QSMRim-Net: Imbalance-aware learning
for identication of chronic active multiple sclerosis lesions on quantitative
susceptibility maps. NeuroImage Clin. 34, 102979 https://doi.org/10.1016/j.
nicl.2022.102979.
F. La Rosa et al.