Content uploaded by Donnie Cameron
Author content
All content in this area was uploaded by Donnie Cameron on Jun 13, 2017
Content may be subject to copyright.
1 3
Magn Reson Mater Phy
DOI 10.1007/s10334-017-0630-3
RESEARCH ARTICLE
Identification of myocardial diffuse fibrosis by 11 heartbeat
MOLLI T1 mapping: averaging to improve precision
and correlation with collagen volume fraction
Vassilios S. Vassiliou1,2,3 · Katharina Wassilew4 · Donnie Cameron1 · Ee Ling Heng2,3 · Evangelia Nyktari2 ·
George Asimakopoulos2 · Anthony de Souza2 · Shivraman Giri5 · Iain Pierce2,3 · Andrew Jabbour6 · David Firmin2,3 ·
Michael Frenneaux1 · Peter Gatehouse2,3 · Dudley J. Pennell2,3 · Sanjay K. Prasad2,3
Received: 10 February 2017 / Revised: 4 May 2017 / Accepted: 24 May 2017
© The Author(s) 2017. This article is an open access publication
myocardial diffuse fibrosis assessment were established
with incremental inclusion of imaging by averaging of the
basal and mid-myocardial left ventricular levels, and each
model was assessed for precision and correlation with col-
lagen volume fraction.
Results A model using 11 heart beat MOLLI imaging of
two basal and two mid ventricular level averaged T1 maps
provided improved precision (Intraclass correlation 0.93 vs
0.84) and correlation with histology (R2 = 0.83 vs 0.36) for
diffuse fibrosis compared to a single mid-ventricular level
alone. ECV was more precise and correlated better than
native T1 mapping.
Conclusion T1 mapping sequences with repeated averag-
ing could be considered for applications of 11 heartbeat
MOLLI, especially when small changes in native T1/ECV
might affect clinical management.
Keywords T1 mapping · MOLLI · Correlation with
collagen volume fraction · Precision · Extracellular
volume · Gadolinium
Introduction
The longitudinal relaxation time, T1, of the myocardium
is regarded as a useful imaging biomarker, as it can
change with cardiac pathology and is known to be associ-
ated with functional capacity and mortality [1–5]. Tradi-
tionally, late gadolinium enhancement (LGE) cardiovas-
cular magnetic resonance (CMR) exploits changes in T1
following administration of a gadolinium–based contrast
agent (Gd): namely, shortening of T1, which manifests as
bright signal intensity on conventional inversion-recov-
ery gradient echo sequences. This has been used as the
primary tool for identification of focal (or replacement)
Abstract
Objectives Our objectives involved identifying whether
repeated averaging in basal and mid left ventricular myo-
cardial levels improves precision and correlation with col-
lagen volume fraction for 11 heartbeat MOLLI T1 mapping
versus assessment at a single ventricular level.
Materials and methods For assessment of T1 mapping
precision, a cohort of 15 healthy volunteers underwent
two CMR scans on separate days using an 11 heartbeat
MOLLI with a 5(3)3 beat scheme to measure native T1 and
a 4(1)3(1)2 beat post-contrast scheme to measure post-
contrast T1, allowing calculation of partition coefficient
and ECV. To assess correlation of T1 mapping with colla-
gen volume fraction, a separate cohort of ten aortic steno-
sis patients scheduled to undergo surgery underwent one
CMR scan with this 11 heartbeat MOLLI scheme, followed
by intraoperative tru-cut myocardial biopsy. Six models of
* Vassilios S. Vassiliou
v.vassiliou@rbht.nhs.uk
1 Norwich Medical School, University of East Anglia, Bob
Champion Research and Education Building, Norwich
Research Park, Norwich NR4 7UQ, UK
2 CMR Unit and NIHR Cardiovascular Biomedical Research
Unit, Royal Brompton Hospital, Sydney Street, London SW3
6NP, UK
3 Imperial College, National Heart and Lung Institute, London,
UK
4 The Pathology Department, Rigshospitalet, University
Hospital of Copenhagen, Blegdamsvej 9, 2100 Copenhagen,
Denmark
5 Siemens Medical Solutions USA, Inc, Chicago, USA
6 Department of Cardiology, St Vincent’s Hospital,
Darlinghurst, Australia
Magn Reson Mater Phy
1 3
fibrosis, indicative of scar, and has entered clinical rou-
tine for multiple pathologies, including myocardial
infarction and viability [6], cardiomyopathy [2, 7], con-
genital heart disease [8] and valvular heart disease [5, 9].
A limitation of T1-weighted inversion recovery
sequences is that they rely on the nulling of signal in
normal myocardium to highlight concentrations of Gd
in fibrotic areas. In cases of diffuse (interstitial) fibro-
sis, however, where the myocardium can be globally
affected, the myocardial signal may appear isointense
and, hence, lack sensitivity in identifying fibrosis [10].
To address this unmet clinical need and allow identifica-
tion of diffuse myocardial fibrosis, new T1 mapping CMR
sequences have been developed, based on the Modified
Look-Locker inversion recovery (MOLLI) sequence
first described by Messroghli and colleagues [11]. These
allow imaging of extracellular volume fractions as a sur-
rogate for diffuse fibrosis [12–16].
Recent iterations of the MOLLI T1 mapping sequence
acquire a total of 8 or 9 T1-weighted images over 11 heart
beats [14, 17], in a 5b(3b)3b scheme pre-contrast, and a
4b(1b)3b(1b)2b scheme post-contrast, where ‘b’ denotes
beats and the values in brackets indicate pause inter-
vals. These shortened acquisitions enable rapid breath-
held imaging in around 8–12 s, depending on heart-
rate, compared to the 15–20 s required by older MOLLI
sequences, which typically acquired T1-weighted images
in a 3b(3b)3b(3b)5b scheme [11, 18]. While reducing the
number of T1-weighted MOLLI source images from 11
to 8/9 may reduce T1 precision, it also has the distinct
advantage of increased reliability through better patient
breath-holding. Therefore, the newer, faster sequences
have enabled extended applications in patients with
poorer breath-holding ability. However, before a new
sequence is fit for clinical application, assessment of pre-
cision and accuracy is required, benchmarked against the
reference standard of histologically derived quantifica-
tion of diffuse fibrosis.
We examined an 11 heartbeat (11HB) MOLLI proto-
type, by Siemens, with native 5b(3b)3b and post-contrast
4b(1b)3b(1b)2b, which is referred to in the main text as
MOLLI for brevity. This sequence has undergone little
exploratory validation, and thus the aim of this work is to:
(1) explore the effect of six imaging models (models A–F,
Table 1) on precision and correlation with collagen
volume fraction (CVF), which we use as a surrogate
for accuracy; these models use incremental averaging
of left ventricular slices (up to two averages at basal
and up to two at mid-ventricular levels, before and
after Gd) here referred to as “incremental inclusion of
images”, and will be compared with the conventional
use of only one mid-ventricular level slice;
(2) histologically validate the 5b(3b)3b and
4b(1b)3b(1b)2b 11 heart beat MOLLI prototype in
patients who have left ventricular myocardial biopsy at
the time of aortic valve surgery; and
(3) explore the relative precision and correlation with CVF
of native T1 mapping, partition coefficient, and extra-
cellular volume fraction (ECV).
Materials and methods
Population cohorts
Two cohorts of participants were recruited for this work: a
group of healthy volunteers for estimating T1 mapping pre-
cision, and a group of aortic stenosis patients for estimat-
ing T1 mapping correlation with CVF. Written consent was
obtained from all patients and volunteers, and the study
was conducted after local research ethics approval from the
Royal Brompton and Harefield NHS Foundation trust and
in accordance with the Declaration of Helsinki principles
for Medical Research.
Cohort for estimating reproducibility/precision
Healthy volunteers taking no medication and with no
known medical conditions were recruited following adver-
tisements on public notice boards. The volunteers under-
went a detailed health questionnaire and blood screening
to measure haemoglobin, renal function, liver function,
thyroid function, and C-reactive protein. The bloods were
measured in a clinical biochemistry laboratory. Addi-
tionally, blood pressure, heart rate, and temperature were
recorded. All volunteers underwent electrocardiography.
In the event that an abnormality (e.g., hypertension, ECG
abnormalities, renal failure) was detected that could have
affected the results, the participant was excluded.
Table 1 Using the four native and four post-Gd images it was pos-
sible to construct a total of 6 T1 mapping models
The “x” marks the slice location of inclusion of the additional imag-
ing
Basal level Repeated Mid-level Repeated
Model A x
Model B x
Model C x x
Model D x x
Model E x x
Model F x x x x
Magn Reson Mater Phy
1 3
Cohort for estimating correlation with collagen volume
fraction
A cohort of patients with severe aortic stenosis, scheduled
for surgical aortic valve replacement, was recruited from
outpatient clinics or inpatient wards at the Royal Bromp-
ton Hospital, London, UK. The patients were excluded
if they had any contraindications to CMR or if they were
scheduled for surgery for more than one valve. Patients
who were scheduled for concurrent aortic valve replace-
ment and coronary artery by-pass grafting (CABG) were
included. As this was a correlation study, the patients were
not excluded if they had treated hypertension, hyperlipidae-
mia, or diabetes.
Assuming similar correlations to histological CVF to
previous published data [12], we estimated that a minimum
of five patients would be required to detect an R2 correla-
tion of at least 0.8 with 80% power. We opted to double
this number to ten patients, giving us 90% power to detect a
statistically significant correlation assuming that the true R2
correlation was at least 0.6.
Cardiovascular magnetic resonance
All CMR imaging was performed on a 1.5 T scanner (Sie-
mens Avanto, Erlangen, Germany). The quadrature body
coil was used for radiofrequency transmission with 12–18
elements of an anterior and posterior cardiac parallel array
coil for reception. The standard manufacturer wireless-
ECG was used as in routine clinical work.
T1 mapping acquisition
The T1 mapping acquisition protocol was the same for both
volunteers and aortic stenosis patients. Following adjust-
ment of the scanner reference frequency to the predominant
Larmor nuclear magnetic resonance precession frequency
of the signal received from a cuboid volume region over the
left-heart (“volume adjust”, ideally performed end-expira-
tion), MOLLI T1 mapping was applied with the following
parameters: Single-shot balanced steady-state free-preces-
sion images, field-of-view = 360 × 306 mm, slice thick-
ness = 8 mm, alpha pulse flip angle = 35o, 7/8ths partial ky
acquisition with 1/8th zero-filling, a generalised autocali-
brating partially parallel acquisition acceleration factor of
2 with 24 central ky lines fully acquired to obtain receiver
coil profiles, and 1085 Hz/pixel ADC sampling bandwidth.
The T1 weighting was initially TI = 120 ms, incrementing
by 80 ms at the start of each Look-Locker set. Two differ-
ent MOLLI acquisitions were used depending on partici-
pant heart rates: a high-resolution acquisition for heart rates
less than 90 bpm and a low-resolution acquisition for heart
rates greater than 90 bpm. Two different sequences were
required because the high-resolution acquisitions were too
long to fit into the stationary diastole phase for faster heart
rates. The following parameters varied between high-reso-
lution and low-resolution setups, and these differences have
been shown to have negligible impact on T1 values while
maximising image quality [10].
High resolution: TR/TE = 2.6/1.12 ms, 84 acquired PE
lines, single-shot image duration 220 ms, Acquired pixel
size 1.4 mm (FE) × 2.1 mm (PE).
Low resolution: TR/TE = 2.4/1.0 ms, 76 acquired PE
lines, single-shot image duration 182 ms, Acquired pixel
size 1.9 mm (FE) × 2.4 mm (PE).
Further, for the low- and high-resolution MOLLI acqui-
sitions, two different MOLLI sampling schemes were used
pre- and post-contrast [10]: a 5b(3b)3b scheme was used
for native T1 values and a 4b(1b)3b(1b)2b scheme was used
for post-contrast T1.
A total of four native T1 maps were acquired (one basal
and repeated once; one mid-ventricular level and repeated
once). Then, starting fifteen minutes after Gd administra-
tion, post-Gd MOLLI T1 maps were acquired at basal and
mid-ventricular levels (and repeated once at each level, giv-
ing rise to four post-Gd T1 maps) using the 4b(1b)3b(1b)2b
scheme, again with low or high resolution depending on the
heart rate. In total, four native T1 maps and four post-con-
trast T1 maps were obtained. The addition of a complete T1
mapping protocol to a clinical scan increased the total scan
time by about 5–7 min per patient.
To assess interscan reproducibility, the healthy volun-
teers underwent a second CMR scan within 60 days of the
first. To mimic routine clinical practice, the staff undertak-
ing the second scan were unaware of the exact slice posi-
tion used for acquisition in the first scan. The aortic steno-
sis patients underwent a single CMR scan as close to the
day of surgery as possible.
Image post‑processing and analysis
For quantification of LV function, volumes, and T1 values,
dedicated software CMR Tools (Cardiovascular Imaging
Solutions, London, UK, Fig. 1) was used by experienced
(level 3 SCMR) blinded operators. Visual inspection of the
colour maps was undertaken both in-line on the Siemens
platform and off-line using CMR 42 (Circle CVI, Calgary,
Canada, Fig. 2).
MOLLI T1 maps were generated automatically in-line
on the scanner workstation, following motion correc-
tion and pixel-by-pixel curve fitting with Look-Locker
correction. DICOM images were anonymised and maps
were analysed in a blinded fashion by two observers
to evaluate interstudy and interobserver variability. A
Magn Reson Mater Phy
1 3
third observer adjudicated values of >5% relative differ-
ence. Native and post-Gd MOLLI source images were
inspected and a quality score from 1 to 5, with five being
excellent and 1 very poor, was recorded for respiratory
drift, cardiac motion artefact, and appropriate cardiac
triggering of each image, before proceeding with meas-
urements on the derived T1-maps. Only patients scoring
≥4 in all categories were subsequently included in the
analysis.
Regions of interest (ROIs) were drawn on the T1 maps
(Fig. 1) within the mid-wall of the septum, with care
being taken to avoid partial-volume or cardiac-motion-
related blurring of the myocardium-blood boundary,
where blood signal contaminates myocardial data. A sec-
ond set of ROIs were drawn in the blood, avoiding myo-
cardium and papillary muscles. A standardised approach
was undertaken in all patients with regards to the location
of the ROI, and all areas were included, even if they were
subsequently shown to have LGE. This had been decided
a priori, as in patients where only native T1 mapping is
undertaken, one does not have the subsequent benefit of
knowing where LGE is present, and therefore our proto-
col was standardised in this way for all the patients.
Once native and post-Gd myocardial and blood val-
ues and haematocrit were obtained, extracellular volumes
(ECVs) were calculated as shown in Eq. (1):
where T1 myopost and T1 myopre are the T1 values of post-
contrast and native myocardium, and T1 bloodpost and
T1 bloodpre are the T1 values of post-contrast and native
blood, respectively.
Myocardial biopsy: procedure and analysis
All aortic stenosis patients included underwent success-
ful myocardial biopsy intraoperatively using the follow-
ing standardised protocol: Once the chest was opened
and cardiopulmonary bypass was established, a long
near-transmural left ventricular “tru-cut” biopsy was
taken. The aim was to undertake multiple biopsies pro-
vided that patient safety was not compromised. Upon
completion of the biopsy, the myocardium was sutured,
(1)
ECV
=(1−haematocrit)∗
1
T1myopost −1
T1myopre
1
T1bloodpost −1
T1bloodpre
Fig. 1 Representative short axis images from a healthy volunteer
with native T1 maps (a–d) and post gad T1 maps (e–h). One basal and
one mid-ventricular level were selected and imaging at each level was
repeated both before and after the administration of Gd (a, b native
T1 basal; e, f post-Gd basal; c, d native T1 mid-ventricular level, g, h,
mid-ventricular level post-Gd). Regions of interest (ROI) were drawn
in the myocardium and blood
Magn Reson Mater Phy
1 3
if required, and surgery continued as normal. The myo-
cardial biopsies were immediately fixed in warm buff-
ered 4% formalin. Histological analysis of fibrosis was
undertaken on 3-µm-thick sections from formalin-fixed,
paraffin-embedded endomyocardial biopsies using a spe-
cial collagen stain (Picrosirius red). Quantification was
performed using standardised semiautomatic image anal-
ysis software (Nikon Advanced Research NIS Elements
imaging software, NIS elements AR 4.10.02, Nikon) on
images of 14 randomly chosen consecutive high-power
fields (×200 magnification), which were obtained with a
Nikon Eclipse E400 light projection microscope (Nikon,
Minato, Tokyo, Japan). The 14 high power fields equaled
1 mm2. Endocardium, subendocardial fibrosis and suben-
docardial fat, procedure-related optically empty spaces
and epicardial fat were eliminated from analysis. Sub-
sequent image analysis was performed to determine the
level of fibrosis including reactive, band-like perimysial
collagen depositions and perivascular fibrosis, defined as
collagen surrounding arterioles. Fibrosis was calculated
as the collagen volume fraction (%) per square millim-
eter, as previously described [19].
Statistical analysis
The CMR native T1 values, partition coefficients, and
ECVs were compared to the histologically identified col-
lagen volume fraction (CVF). Intraclass correlation coeffi-
cients (ICC) were calculated using ‘R’ (R Foundation for
Statistical Computing, Vienna, Austria) and used to assess
agreement in the healthy volunteer ‘precision’ cohort. For
the aortic stenosis ‘correlation with CVF’ cohort, graph-
plotting and analysis was again undertaken using the statis-
tics package ‘R’. The six models, A–F, described for CMR
estimation of myocardial diffuse fibrosis were compared to
identify the most reproducible model (from the volunteer
precision cohort) and most accurate/correlating with CVF
(from the aortic stenosis biopsy cohort). There is no pub-
lished method of incorporating both precision and accuracy
(via correlation with CVF) in one single measurement for
such imaging work. We therefore proposed and calculated
the product of the scan-rescan R2 values for the volunteer
(precision) and ECV-histological fibrosis R2 for aortic ste-
nosis biopsy patients (correlation with CVF) as a single
combined measure of correlation with CVF and precision
Fig. 2 a showing a basal slice
with native T1 mapping using
the high resolution 5(3)3 beat
sequence as the heart rate was
58 bpm. b shows the post-Gd
image using a high resolution
4(1)3(1)2 beat sequence. c
shows the native T1 mapping
and d the post-Gd contrast T1
mapping using the same high
resolution sequence as in (a,
b) respectively, but at the mid-
ventricular level
Magn Reson Mater Phy
1 3
for this imaging modality, described in this manuscript as
Product of R2.
Results
Reproducibility‑precision: the volunteer cohort
Fifteen healthy volunteers, eight males, mean
age = 31 ± 5 years, were recruited and the reproducibility of
their corresponding native T1 maps, post Gd T1 maps, parti-
tion coefficient and ECV results are shown in Table 2. All vol-
unteers had an HR < 90 bpm for both scans, and; therefore,
the higher resolution sequence was used. Participants toler-
ated both scans well with no complications and excellent T1
mapping images were obtained. No volunteer was excluded
due to respiratory drift, cardiac motion artefact, or inappro-
priate cardiac triggering. There was no statistically significant
difference between the two scans with regards to heart rate,
blood pressure, or haematocrit (p = 0.27, p = 0.19, p = 0.89,
respectively). Both interstudy T1 and interstudy ECV dif-
ferences showed excellent results (Bland–Altman shown in
Fig. 3 for native T1 and Fig. 4 for ECV, respectively).
Both intra- and interobserver variability were excellent.
Intraobserver ICCs were: 0.995 for native T1 (p < 0.001),
0.996 for partition coefficient (p < 0.001) and 0.999 for
Table 2 Showing interscan
reproducibility
The six models A–F incorporating increasing levels and averaging of myocardial T1 maps are shown.
Model C–F showed the best extracellular volume (ECV) reproducibility
ICC intraclass correlation
Interstudy reproducibility
Basal Repeat Mid Repeat Native T1
mapping
Post Gd mapping Partition coefficient ECV
ICC
p value
ICC
p value
ICC
p value
ICC
p value
Model A x 0.90
P < 0.001
0.72
p = 0.014
0.53
p = 0.098
0.84
p = 0.001
Model B x 0.84
p = 0.001
0.68
p = 0.018
0.4
p = 0.17
0.75
p = 0.006
Model C x x 0.88
p < 0.001
0.82
p = 0.002
0.81
p = 0.002
0.94
p < 0.001
Model D x x 0.80
p = 0.001
0.75
p = 0.006
0.66
p = 0.024
0.88
p < 0.001
Model E x x 0.90
p < 0.001
0.71
p = 0.013
0.54
p = 0.08
0.88
p < 0.001
Model F x x x x 0.88
p < 0.001
0.79
p = 0.003
0.81
p = 0.002
0.93
P < 0.001
Fig. 3 Bland–Altman analysis of T1, including mean difference, correlation, and distribution of differences plots. Mean difference = 11.5 ms
(red line on difference plot) with limits of agreement at −21 and 43 ms; the line of zero difference (dashed) is within the limits of agreement
Magn Reson Mater Phy
1 3
ECV (p < 0.001). Interobserver ICCs were: 0.993 for native
T1 (p < 0.001), 0.985 for partition coefficient (p < 0.001),
and 0.997 for ECV (p < 0.001).
Correlation with collagen volume fraction: The aortic
stenosis cohort
Ten consecutive patients with symptomatic severe aor-
tic stenosis scheduled for surgical valve replacement
underwent CMR and intraoperative myocardial biop-
sies: eight male, mean age = 71 ± 10 years, and three
with significant major epicardial coronary artery disease
(defined as disease >50% of lumen diameter) requiring
CABG at the time of the operation. Of these patients,
five received treatment for hypertension and their blood
pressure was well controlled at the time of the CMR
(BP < 140/90 mmHg), four had treated hyperlipidaemia,
two had chronic obstructive pulmonary disease requir-
ing bronchodilators and two had prior distant history of
neoplasia (breast/bladder). Of the three patients requiring
CABG, one had 2-vessel CABG (left anterior descending
(LAD) and circumflex) and the other two had single ves-
sel CABG (one of the LAD and one of the intermediate
coronary artery respectively). No patients suffered from
diabetes. All patients tolerated the CMR well and the T1
mapping images were free from disqualifying artefacts
and were diagnostic in all participants. In one patient the
initial native T1 maps needed repeating in view of poor
breathholding. This was appreciated during the scan, and
therefore was repeated before proceeding to Gd adminis-
tration. The repeat maps were of acceptable quality with
no apparent respiratory drift, cardiac motion artefact, or
inappropriate cardiac triggering. All patients had api-
cal–lateral wall biopsies taken intraoperatively and no
patients had biopsies from all three basal, mid, and apical
levels due to surgical concerns. One patient had to be
taken back to theatre within 24 h from surgery for bleed-
ing, which may have been related to the apical myocar-
dial biopsy sampling. The patient made a good recovery
and discharged home within five days with no further
sequela. No other side-effects relating to the myocardial
biopsies were seen. Therefore, the basal and/or mid-myo-
cardial T1 values were taken as a surrogate for a global T1
value and were correlated with histology from the apical–
lateral wall. Microscopic views of the sections prepared
from the biopsies are shown in Fig. 5.
The relative correlation with CVF of native T1 maps,
partition coefficient, and ECV were assessed using histol-
ogy. The histologically derived CVF was plotted against
each of these parameters, as shown in Table 3. Native T1
values from model F (which had shown best correlation)
showed a moderate, statistically-significant correlation with
the histologically obtained fibrosis burden, with R2 = 0.42,
and p = 0.046. However, the partition coefficient and
ECV both showed strong correlations with histology, with
R2 = 0.81 and R2 = 0.83, and these both were statistically
significant, with p < 0.001 (Table 3; Fig. 6).
Having established that ECV was the most accurate
imaging parameter from T1 mapping, we further inves-
tigated the incremental addition of more myocardial
regions and averaged slices. When only one mid-level T1-
map was used (i.e., Model A), the correlation with histol-
ogy was only modest, and was not significant (R2 = 0.36,
p = 0.068), similar to data from published studies that used
only one mid-level T1-map [14, 20]. Increasing the num-
ber of T1 maps, and including the basal level as well as the
mid-ventricular-level in this averaging improved the corre-
lation significantly: R2 = 0.72, p = 0.002 when the aver-
age of a single basal and single mid-level image was used
for ECV (Model E) and importantly R2 = 0.83, p < 0.001
Fig. 4 Bland–Altman analysis of ECV, including mean difference, correlation, and distribution of differences plots. Mean difference = 0.003
(red line on difference plot) with limits of agreement at −0.024 and 0.029; the line of zero difference (dashed) is within the limits of agreement
Magn Reson Mater Phy
1 3
when acquisition was repeated at both basal and mid-levels
(Model F) as shown in Table 3 and Fig. 7.
Combining precision and correlation with collagen
volume fraction
To allow us to clarify the relative incremental addition
of imaging for each model combining both accuracy (via
correlation with CVF) and precision we sought a single
parameter for each model to incorporate both correlation
with CVF and precision. However, such a single meas-
ure has not been described for T1 mapping. We therefore
devised a new model, incorporating the product of the R2
from the precision cohort and R2 from the ‘correlation with
CVF’ cohort as a unique parameter for each model. The
results are shown in Table 4, and indicate that the inclusion
of a repeated basal image and a repeated mid-image, giving
a total of four native and four post Gd T1 maps to calculate
ECV, was the most precise method with the best correlation
with CVF.
Discussion
Any new CMR sequence needs to show both good accu-
racy and precision to become clinically relevant. In this
work, we used two cohorts to establish precision and
correlation with CVF (as a surrogate for accuracy) for
an as yet non-validated T1 mapping sequence. Further-
more, this is the first time a study has reported a model
of incremental image acquisition and averaging of basal
and mid-ventricular T1-measurements as a means of
obtaining an average of the global T1-mapping ECV. The
volunteer cohort demonstrated that four models of slice
Fig. 5 Samples from the intraoperative myocardial biopsies stained
with Picrosirius red. a Shows fibrous septae, which show perivascu-
lar collagen to support mural arteries. There is no scarring, but only
a slight increase in collagen fibres surrounding each cardiomyocyte
(light red). b Shows interstitial fibrosis only, each cardiomyocyte
is supported by a thin collagen layer (light red), there is only focal
perivascular(pericapillary–capillary encircled) increase in collagen
fibers (dark red area annotated). c Shows an annotated red area, quali-
fying as scar, as the dimension of the area exceeds double the diam-
eter of the adjacent hypertrophic cardiomyocyte indicated with ⟷
Table 3 showing the correlation of extracellular volume (ECV) by each model with histological collagen volume fraction (CVF) from apical–
lateral histology
It can be argued that in view of other myocardial pathologies co-existing with aortic stenosis such as oedema or ischaemia, CVF might not be
the gold standard for fibrosis comparison. Therefore, this table should be interpreted in the appropriate clinical context. In patients with aortic
stenosis, it would appear that ECV derived from the basal slice correlated better with CVF than the mid-ventricular level, but a combination of
both the basal and mid had the best correlation. These results would suggest that further research is warranted to show whether imaging of basal
and mid-ventricular level should be routinely undertaken in patients with aortic stenosis
Correlation of ECV with histological fibrosis
Basal level Basal repeated Mid-level Mid-level repeated ECV (%) R2 with histology p value
Model A x 27.47 0.36 0.068
Model B x 26.67 0.77 0.001
Model C x x 26.86 0.49 0.025
Model D x x 26.79 0.81 <0.001
Model E x x 27.07 0.72 0.002
Model F x x x x 26.83 0.83 <0.001
Magn Reson Mater Phy
1 3
acquisition and averaging showed an excellent interscan
correlation. Model A native T1 mapping (one native T1
mid-ventricular level slice), Model E native T1 mapping
(average of one basal and one mid-level slice native T1),
Model C ECV (average of ECV of two mid-level slices),
and Model F ECV (average ECV of two basal and two
mid-slices) all showed excellent interscan reproducibility,
with R2 ≥ 0.90.
The correlation with CVF was assessed using a cohort
of ten patients with aortic stenosis scheduled for surgical
valve replacement who had CMR using our model of incre-
mental addition of imaging acquisitions and intraoperative
biopsies. This confirmed that in aortic stenosis patients,
measuring ECV in the mid-ventricular wall alone (Model
A) showed only a trend towards weak correlation with
histological collagen volume fraction (CVF), R2 = 0.36,
p = 0.068. A model utilising the basal level alone (Model
B) showed good correlation with histology (R2 = 0.77,
p = 0.001), while a model based on one basal level being
imaged twice and averaged (Model D) showed better cor-
relation (R2 = 0.81, p < 0.001). The best correlation was
achieved by a model having one basal and one mid-ventric-
ular slice being imaged twice each and averaged (Model
F, R2 = 0.83, p < 0.001). Incorporating both precision
and correlation with CVF parameters into the Product R2,
showed that ECV calculated from Model F was the most
accurate and precise. We speculate that the improved corre-
lation and precision associated with Model F may relate to
the incorporation of further areas of the myocardium (both
basal and mid-ventricular level) and the repeat scan may
serve to average out any potential errors associated with
imaging a single slice.
This work therefore supports the following conclu-
sions: firstly, the 11 heart beat MOLLI prototype incorpo-
rating native 5b(3b)3b and post-contrast 4b(1b)3b(1b)2b
schemes, was well tolerated by volunteers and patients.
Secondly, when isolated native T1 mapping was used
(without post-contrast T1 maps), it is reproducible in
healthy volunteers, especially when Model A (one mid-
slice), Model C (one mid-slice repeated), Model E (one
basal and one mid-slice), or Model F (one basal and one
mid-ventricular level, both repeated) are used. Repeat-
ing native T1 maps does not appear to offer additional
improvement in precision, as Model A performed at least
as well as Models C, E and F. Thirdly, both partition coef-
ficient and ECV have been shown to be more accurate
than native T1 maps alone. Therefore, we propose that
post-Gd imaging (with or without haematocrit sampling)
should be routinely undertaken when looking for an esti-
mation of diffuse fibrosis in patients. Fourthly, a sin-
gle mid-ventricular ECV measurement appears to show
poorer correlation with CVF in patients with aortic ste-
nosis (as shown by Model A). Providing additional image
repeats, such as model C, E or F for example, appear to
show improved correlation with CVF. Although our study
is not without its limitations, particularly regarding the
use of the CVF as the reference standard, further work
could consider whether the additional improvement in
correlation provided by models C, E, and F, for example,
might be of clinical importance. Finally, the most accu-
rate and precise model is Model F, whereby imaging of
one basal and one mid-ventricular level slice, with both
being repeated once, is performed both in the native state
and after Gd administration.
Fig. 6 The agreement comparing (apical) histological CVF against
ECV (a), native T1 mapping (b), and partition coefficient (c). There
was good agreement between CVF and all the imaging parameters;
however, partition coefficient and ECV performed considerably better
than native T1 mapping alone
Magn Reson Mater Phy
1 3
Fig. 7 a Showing the correlation of (apical–lateral) histological col-
lagen volume fractions (CVF) with ECV calculated from CMR at a
single mid-ventricular level, corresponding to Model A and showing
only mild non-significant correlation. b Showing correlation of his-
tological fibrosis with a single basal level (Model B), showing that
for this pathology the correlation increases and has now become sig-
nificant. c Showing correlation between the average of one mid-level
and one basal level (Model E) with histology showing a significant
correlation. d Represents Model F with two basal and two mid-
slices showing that this model demonstrated the strongest correlation
(R2 = 0.83, p < 0.001). This work confirms that including incremen-
tally more levels imaged improves accuracy
Table 4 Combining accuracy
and precision in a single
measurement, the Product of R2
indicating that Model F offers
the best imaging protocol (basal
and repeat, mid-ventricular level
and repeat) for optimal accuracy
and precision
ECV extracellular volume
Basal Repeat Mid Repeat Volunteer interscan R2ECV vs histology R2Product of R2
Model A x 0.759 0.356 0.27
Model B x 0.722 0.773 0.56
Model C x x 0.808 0.486 0.39
Model D x x 0.637 0.809 0.52
Model E x x 0.623 0.715 0.45
Model F x x x x 0.778 0.829 0.65
Magn Reson Mater Phy
1 3
We therefore recommend that this should be consid-
ered if there is available scanner time and the patients
can tolerate it. However, increasing scanner time (even by
5–7 min) might not be possible in many busy CMR units.
As such, undertaking repeats of basal and mid-ventricular
levels might not be an option. Nonetheless, even in busy
units it should be possible to undertake repeat imaging of
a single level, which will only add two breathholds to the
scan time. Therefore, if a unit undertakes a mid-ventricu-
lar slice T1 mapping routinely in the native and post-con-
trast state, we recommend that imaging at the same level
is repeated once. This will provide additional improve-
ment in the correlation-precision, but also safeguard
against potential poor breathholding, poor coregistration,
or motion correction in one of the two acquisitions. Such
a practical compromise will only increase scanner time by
about 30 s. It should be noted that this relation only holds
true for the 11HB sequence we have tested, and other
MOLLI sequences might not behave in a similar way.
This study has certain limitations. Firstly, our samples for
both cohorts were small; however, this is in line with other
recently published studies, which also use small (fewer
than ten patients) datasets for histological validation [21,
22]. Secondly, we used healthy volunteers for the interscan
reproducibility, as this was practically easier to achieve. As
aortic stenosis patients can be more breathless and might
have difficulties with breathholding, it is possible that the
reproducibility might not be as good as for healthy controls.
However, recently published work [23, 24] has shown that
reproducibility in aortic stenosis was excellent, and there-
fore, we would not have anticipated different results to our
controls, as T1 mapping in our aortic stenosis patients was
of very good quality. Furthermore, we elected not to include
apical T1 mapping in any of our T1 mapping strategies, as
this region is known to suffer increased partial-volume
artefact at the endocardial and epicardial borders, which
are not perpendicular to the image slice in the apical short-
axis region [25]. Moreover, we did not undertake intra-
study reproducibility which could have provided additional
information. We felt that intrascan reproducibility for ECV
would be difficult to interpret as by default the post-Gd T1
maps would be taken at different times. In addition, our
myocardial biopsies were taken from the apico-lateral wall.
Nonetheless, earlier studies have shown that, unlike replace-
ment fibrosis, where there is a significant difference between
basal and apical areas [26], interstitial fibrosis tends to fol-
low a more homogenous pattern [23, 27] in patients with
aortic stenosis. Therefore, we feel that our results are valid
for the apico-lateral wall biopsies to represent global diffuse
fibrosis measurement in these patients. Finally, it is appreci-
ated that other pathological processes might affect T1/ECV
in aortic stenosis including oedema [25] and ischaemia [28,
29]; therefore, the correlation between T1/ECV and CVF
should be regarded as an association rather than causative,
as there could be multiple aetiologies that might increase T1/
ECV.
Conclusion
This work was the first to utilise T1 mapping strategies with
increasingly inclusive additional imaging for averaging the
septal ECV measurements, going from one slice imaged
to a total of four images (one basal repeated and one mid
repeated) to improve both the correlation with CVF and
the precision of the T1-measurement (both before and after
Gd), comparing histological CVF estimation of fibrosis
to the T1-mapping outcomes of native T1 maps, partition
coefficient and ECV. We have shown that the 11 heartbeat
MOLLI sequence used in this research, native 5b(3b)3b
with post-contrast 4b(1b)3b(1b)2b, was well-tolerated in
both volunteers and patients with aortic stenosis. Further-
more, it was both accurate and precise, particularly as incre-
mental imaging of additional basal and mid-ventricular slice
images are included. Finally, this work also confirms that
partition coefficient and ECV can be accurately used to cor-
relate with histological CVF estimation of diffuse fibrosis.
Acknowledgements The authors would like to acknowledge the
assistance of Mr Alex Bowman, Histopathology Core Facility Labora-
tory Manager, Biomedical Research Unit, Royal Brompton Hospital,
London with the preparation of the histology blocks.
Funding This work was supported by the National Institute of Health
Research Cardiovascular Biomedical Research Unit at the Royal
Brompton Hospital and Imperial College, London and Rosetrees
Trust charity. Dr Heng was supported by a British Heart Foundation
Clinical Research Training fellowship (FS/13/76/30477).
Author contributions VV, PG and SKP conceived and designed
the study, VV wrote the paper. KW, ELH, EN, AJ provided detailed
review of the study design and undertook experimental work and
analysis. DC, SG, IP, DF, MF, DJP provided detailed review of the
study design, results and conclusions. GA, AdS contributed to the
study design, undertook experimental work and reviewed results and
conclusions. All authors read, made critical revisions and approved
the final manuscript.
Compliance with ethical standards
Conflict of interest Professor Pennell is a stockholder and director of
Cardiovascular Imaging Solutions and received research support from
Siemens. Professor Frenneaux has received funding from Medtronic
for investigator-initiated research. Dr Prasad has received speaker’s
fees from Bayer. Dr Giri is an employee of Siemens Healthcare. The
remaining authors have no competing interests.
Ethical approval All procedures performed in studies involving
human participants were in accordance with the ethical standards of
the institutional research committee and with the 1964 Declaration of
Helsinki and its later amendments or comparable ethical standards.
Magn Reson Mater Phy
1 3
Informed consent Informed consent was obtained from all individual
participants included in the study.
Open Access This article is distributed under the terms of the Crea-
tive Commons Attribution 4.0 International License (http://crea-
tivecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided you give
appropriate credit to the original author(s) and the source, provide a
link to the Creative Commons license, and indicate if changes were
made.
References
1. Dweck MR, Joshi S, Murigu T, Alpendurada F, Jabbour A,
Melina G, Banya W, Gulati A, Roussin I, Raza S, Prasad NA,
Wage R, Quarto C, Angeloni E, Refice S, Sheppard M, Cook
SA, Kilner PJ, Pennell DJ, Newby DE, Mohiaddin RH, Pepper
J, Prasad SK (2011) Midwall fibrosis is an independent predictor
of mortality in patients with aortic stenosis. J Am Coll Cardiol
58:1271–1279
2. Gulati A, Jabbour A, Ismail TF, Guha K, Khwaja J, Raza S,
Morarji K, Brown TDH, Ismail NA, Dweck MR, Di Pietro E,
Roughton M, Wage R, Daryani Y, O’Hanlon R, Sheppard MN,
Alpendurada F, Lyon AR, Cook SA, Cowie MR, Assomull RG,
Pennell DJ, Prasad SK (2013) Association of fibrosis with mor-
tality and sudden cardiac death in patients with nonischemic
dilated cardiomyopathy. JAMA 309:896–908
3. Wong TC, Piehler K, Meier CG, Testa SM, Klock AM, Aneizi
AA, Shakesprere J, Kellman P, Shroff SG, Schwartzman DS,
Mulukutla SR, Simon MA, Schelbert EB (2012) Association
between extracellular matrix expansion quantified by cardiovas-
cular magnetic resonance and short-term mortality. Circulation
126:1206–1216
4. Wong TC, Piehler KM, Kang IA, Kadakkal A, Kellman P,
Schwartzman DS, Mulukutla SR, Simon MA, Shroff SG, Kuller
LH, Schelbert EB (2014) Myocardial extracellular volume
fraction quantified by cardiovascular magnetic resonance is
increased in diabetes and associated with mortality and incident
heart failure admission. Eur Heart J 35:657–664
5. Vassiliou VS, Perperoglou A, Raphael CE, Joshi S, Malley T,
Everett R, Halliday B, Pennell DJ, Prasad SK (2017) Midwall
fibrosis and 5-year outcome in patients with moderate and severe
aortic stenosis. J Am Coll Cardiol 69:1755–1756
6. Pennell DJ, Sechtem UP, Higgins CB, Manning WJ, Pohost GM,
Rademakers FE, van Rossum AC, Shaw LJ, Yucel EK (2004)
Clinical indications for cardiovascular magnetic resonance
(CMR): consensus Panel report. Eur Heart J 25:1940–1965
7. Ismail TF, Jabbour A, Gulati A, Mallorie A, Raza S, Cowling
TE, Das B, Khwaja J, Alpendurada FD, Wage R, Roughton M,
McKenna WJ, Moon JC, Varnava A, Shakespeare C, Cowie MR,
Cook SA, Elliott P, O’Hanlon R, Pennell DJ, Prasad SK (2014)
Role of late gadolinium enhancement cardiovascular magnetic
resonance in the risk stratification of hypertrophic cardiomyopa-
thy. Heart 100:1851–1858
8. Babu-Narayan SV, Kilner PJ, Li W, Moon JC, Goktekin O,
Davlouros PA, Khan M, Ho SY, Pennell DJ, Gatzoulis MA
(2006) Ventricular fibrosis suggested by cardiovascular mag-
netic resonance in adults with repaired tetralogy of fallot and
its relationship to adverse markers of clinical outcome. Circu-
lation 113:405–414
9. Barone-Rochette G, Piérard S, Meester De, de Ravenstein C,
Seldrum S, Melchior J, Maes F, Pouleur A-C, Vancraeynest D,
Pasquet A, Vanoverschelde J-L, Gerber BL (2014) Prognostic
significance of LGE by CMR in aortic stenosis patients under-
going valve replacement. J Am Coll Cardiol 64:144–154
10. Kellman P, Hansen MS (2014) T1-mapping in the heart: accu-
racy and precision. J Cardiovasc Magn Reson 16:2
11. Messroghli DR, Radjenovic A, Kozerke S, Higgins DM,
Sivananthan MU, Ridgway JP (2004) Modified Look-Locker
inversion recovery (MOLLI) for high-resolution T1 mapping
of the heart. Magn Reson Med 52:141–146
12. Flett AS, Hayward MP, Ashworth MT, Hansen MS, Taylor
AM, Elliott PM, McGregor C, Moon JC (2010) Equilibrium
contrast cardiovascular magnetic resonance for the measure-
ment of diffuse myocardial fibrosis: preliminary validation in
humans. Circulation 122:138–144
13. Roujol S, Weingärtner S, Foppa M, Chow K, Kawaji K, Ngo
LH, Kellman P, Manning WJ, Thompson RB, Nezafat R (2014)
Accuracy, precision, and reproducibility of four T1 mapping
sequences: a head-to-head comparison of MOLLI, ShMOLLI,
SASHA, and SAPPHIRE. Radiology 272:683–689
14. Bull S, White SK, Piechnik SK, Flett AS, Ferreira VM, Lou-
don M, Francis JM, Karamitsos TD, Prendergast BD, Robson
MD, Neubauer S, Moon JC, Myerson SG (2013) Human non-
contrast T1 values and correlation with histology in diffuse
fibrosis. Heart 99:932–937
15. Dabir D, Child N, Kalra A, Rogers T, Gebker R, Jabbour A,
Plein S, Yu C-Y, Otton J, Kidambi A, McDiarmid A, Broad-
bent D, Higgins DM, Schnackenburg B, Foote L, Cummins C,
Nagel E, Puntmann VO (2014) Reference values for healthy
human myocardium using a T1 mapping methodology: results
from the International T1 Multicenter cardiovascular magnetic
resonance study. J Cardiovasc Magn Reson 16:69
16. Liu C-Y, Chang Liu Y, Wu C, Armstrong A, Volpe GJ, van
der Geest RJ, Liu Y, Hundley WG, Gomes AS, Liu S, Nacif
M, Bluemke DA, Lima JAC (2013) Evaluation of age related
interstitial myocardial fibrosis with cardiac magnetic reso-
nance contrast-enhanced T1 mapping in the Multi-ethnic Study
of Atherosclerosis (MESA). J Am Coll Cardiol 62:1280–1287
17. White SK, Sado DM, Fontana M, Banypersad SM, Maestrini
V, Flett AS, Piechnik SK, Robson MD, Hausenloy DJ, Sheikh
AM, Hawkins PN, Moon JC (2013) T1 mapping for myocar-
dial extracellular volume measurement by CMR: bolus only
versus primed infusion technique. JACC Cardiovasc Imaging
6:955–962
18. Messroghli DR, Plein S, Higgins DM, Walters K, Jones TR,
Ridgway JP, Sivananthan MU (2006) Human myocardium: sin-
gle-breath-hold MR T1 mapping with high spatial resolution–
reproducibility study. Radiology 238:1004–1012
19. Wassilew K, Terziev D, Wassilew G, Fitzl G, Frölich K, Kan-
dolf R, Fried A (2016) Ultrastructural morphometric findings of
cardiomyocytes in patients with impaired ventricular function—
a comparative clinicopathological study. Cardiovasc Pathol
25:25–32
20. Iles L, Pfluger H, Phrommintikul A, Cherayath J, Aksit P, Gupta
SN, Kaye DM, Taylor AJ (2008) Evaluation of diffuse myocar-
dial fibrosis in heart failure with cardiac magnetic resonance con-
trast-enhanced T1 mapping. J Am Coll Cardiol 52:1574–1580
21. Ravenstein CDM, Bouzin C, Lazam S, Boulif J, Amzulescu M,
Melchior J, Pasquet A, Vancraeynest D, Pouleur A-C, Vanover-
schelde J-LJ, Gerber BL (2015) Histological Validation of meas-
urement of diffuse interstitial myocardial fibrosis by myocardial
extravascular volume fraction from Modified Look-Locker imag-
ing (MOLLI) T1 mapping at 3 T. J Cardiovasc Magn Reson 17:48
22. Chin CWL, Everett RJ, Kwiecinski J, Vesey AT, Yeung E, Esson
G, Jenkins W, Koo M, Mirsadraee S, White AC, Japp AG, Prasad
SK, Semple S, Newby DE, Dweck MR (2016) Myocardial fibro-
sis and cardiac decompensation in aortic stenosis. JACC Cardio-
vasc Imaging. doi:10.1016/j.jcmg.2016.10.007
Magn Reson Mater Phy
1 3
23. Chin CWL, Semple S, Malley T, White AC, Mirsadraee S, Weale
PJ, Prasad S, Newby DE, Dweck MR (2013) Optimization and
comparison of myocardial T1 techniques at 3T in patients with
aortic stenosis. Eur Heart J Cardiovasc Imaging 15:556–565
24. Singh A, Horsfield MA, Bekele S, Khan JN, Greiser A, McCann
GP (2015) Myocardial T1 and extracellular volume fraction
measurement in asymptomatic patients with aortic stenosis:
reproducibility and comparison with age-matched controls. Eur
Hear J 16:763–770
25. Moon JC, Messroghli DR, Kellman P, Piechnik SK, Robson MD,
Ugander M, Gatehouse PD, Arai AE, Friedrich MG, Neubauer S,
Schulz-Menger J, Schelbert EB (2013) Myocardial T1 mapping
and extracellular volume quantification: a Society for Cardiovas-
cular Magnetic Resonance (SCMR) and CMR Working Group of
the European Society of Cardiology consensus statement. J Car-
diovasc Magn Reson 15:92
26. Hoffmann R, Altiok E, Friedman Z, Becker M, Frick M
(2014) Myocardial deformation imaging by two-dimensional
speckle-tracking echocardiography in comparison to late
gadolinium enhancement cardiac magnetic resonance for analy-
sis of myocardial fibrosis in severe aortic stenosis. Am J Cardiol
114:1083–1088
27. Stuckey DJ, McSweeney SJ, Thin MZ, Habib J, Price AN, Fie-
dler LR, Gsell W, Prasad SK, Schneider MD (2014) T1 mapping
detects pharmacological retardation of diffuse cardiac fibrosis in
mouse pressure-overload hypertrophy. Circ Cardiovasc Imaging
7:240–249
28. Mahmod M, Piechnik SK, Levelt E, Ferreira VM, Francis JM,
Lewis A, Pal N, Dass S, Ashrafian H, Neubauer S, Karamitsos
TD (2014) Adenosine stress native T1 mapping in severe aortic
stenosis: evidence for a role of the intravascular compartment on
myocardial T1 values. J Cardiovasc Magn Reson 16:92
29. Liu A, Wijesurendra RS, Francis JM, Robson MD, Neubauer S,
Piechnik SK, Ferreira VM (2016) Adenosine stress and rest T1
mapping can differentiate between ischemic, infarcted, remote,
and normal myocardium without the need for gadolinium con-
trast agents. JACC Cardiovasc Imaging 9:27–36