Key words CT - thorax - MR imaging
Introduction
Pulmonary emphysema, as a common finding in long-term smokers and a leading cause
of mortality worldwide, is characterized by irreversible destruction of the lung parenchyma
[1 ]
[2 ]. Computed tomography (CT) plays an important role in the visualization and characterization
of pulmonary emphysema, and it has previously been shown that CT findings can predict
mortality in COPD [1 ]
[3 ]. Moreover, the degree of emphysema on CT images can be quantified by computing low
attenuation areas under a specific threshold [4 ]
[5 ]
[6 ].
Magnetic resonance imaging (MRI) allows for radiation-free lung imaging, but the low
proton density of the lung parenchyma, susceptibility artifacts at air-tissue interfaces,
and the vulnerability to respiratory and cardiovascular motion artifacts makes lung
imaging with MRI still challenging [7 ]
[8 ]
[9 ].
Even though a decrease in lung density further complicates image acquisition, several
studies have shown the feasibility of MRI for the assessment of pulmonary ventilation
and emphysema. Most studies report on the use of ultrashort echo time (UTE) imaging,
as well as functional MRI with Fourier decomposition or inhalation of hyperpolarized
noble gases [10 ]
[11 ]
[12 ]
[13 ]
[14 ]
[15 ]
[16 ]
[17 ]. However, especially the latter approach is highly technically demanding.
Besides the aforementioned MRI techniques, it has been shown that conventional structural
T2-weighted MRI using the Periodically Rotated Overlapping ParallEL Lines with Enhanced
Reconstruction (PROPELLER) technique can depict pulmonary nodules and changes in lung
tissue in general [18 ]
[19 ]
[20 ]
[21 ]. Hence, the aim of this study was to evaluate whether T2-weighted PROPELLER imaging
might also be used for the assessment of pulmonary emphysema.
Materials and methods
Study participants
The study population consisted of 224 participants in a lung cancer screening study
comparing low-dose CT (LDCT) and MRI. To be included for the screening, the participants
had to be 50 – 70 years old with a long history of cigarette smoking (at least 15
cigarettes per day for at least 25 years, or at least 10 cigarettes per day for at
least 30 years). Study participants were active smokers or had quit for not more than
10 years. The screening study was approved by the institutional review board and by
the federal agency for radiation protection. Written informed consent was obtained
from all study subjects. For the present study, we retrospectively included 30 participants
with pulmonary emphysema according to LDCT, and 30 participants without emphysema.
Technique
For this study, a transverse T2-weighted sequence using the PROPELLER technique (MultiVane
XD, Philips Healthcare, Best, The Netherlands) as part of our MRI screening protocol
was evaluated. The sequence was acquired on a clinical 1.5 Tesla scanner (Ingenia,
Philips Healthcare, Best, The Netherlands) with an anterior phased-array body coil.
The imaging parameters were as follows: repetition time 2200 – 2500 ms, echo time
60 ms, flip angle 90°, FOV 400 mm, matrix 432 × 432 mm, slice thickness 6 mm, acquisition
time 3:18 min with respiratory gating. Other sequences of the MRI protocol were transverse
T2-weighted STIR (short tau inversion recovery) MVXD, coronal T2-weighted MVXD, transverse
balanced steady-state free precession, and coronal 3D T1-weighted gradient echo, yet
these sequences were not evaluated for this present study.
LDCT was performed on a clinical 128-slice spiral CT scanner (iCT, Philips Healthcare,
Best, The Netherlands) in inspiratory breath-hold with a reconstructed slice thickness
of 2 mm. The tube current-time product was 25 mAs, the tube voltage was 120 kV and
the volume CT dose index was 1.8 mGy, leading to a dose length product of 70 – 90
mGy*cm. All participants underwent LDCT and MRI within the same day or week.
Image analysis
LDCT images of 224 participants of our lung cancer screening program were retrospectively
evaluated for the presence of pulmonary emphysema by a radiologist with 4.5 years
of experience. The window settings of the LDCT datasets for this analysis were window
width 1500 Hounsfield Units (HU) and window level – 700 HU. First, the presence and
severity of emphysema were assessed qualitatively using a three-point scale: 0 = no
emphysema, 1 = moderate emphysema (centrilobular lucencies occupying approximately
> 5 % of a lung zone OR scattered small juxtapleural lucencies), 2 = severe emphysema
(coalescent centrilobular and lobar lucencies including multiple regions OR multiple
mainly large juxtapleural lucencies). Second, morphological patterns of emphysema
were evaluated: 1 = predominantly centrilobular emphysema with scattered, multiple
or coalescent centrilobular lucencies (few paraseptal lucencies may be present), 2 = predominantly
paraseptal emphysema with scattered or multiple paraseptal lucencies (few centrilobular
lucencies may be present), 3 = mixed or advanced destructive emphysema. These definitions
followed the statements of the Fleischner Society published in 2015 [4 ]. Third, automated emphysema analysis was performed using a commercially available
software program (IntelliSpace Portal, Philips Healthcare, Best, The Netherlands)
in order to obtain emphysema indices (EI).
Lung parenchyma was considered emphysematous when it showed attenuation values of
below – 930 HU at inspiration. An emphysema index (EI) was calculated for each LDCT
dataset, defined as the percentage of lung volume with emphysema divided by the total
lung volume. The presence of pulmonary emphysema was defined as an EI of ≥ 6 % or
when multiple lucencies were clearly visible on LDCT. LDCT datasets in which extensive
atelectasis and/or pulmonary infiltrates led to an opacification of approximately
one third of a pulmonary lobe were excluded (n = 10). Subjects for which the automated
analysis of LDCT images did not work appropriately due to noise overlay were also
excluded (n = 27). This was seen in adipose individuals in particular. There are three
reasons for the malfunction of the automated analysis: 1) Automated lung segmentation
from trachea and surrounding tissue failed. 2) Quantitative results showed excessive
emphysema indices that were apparently false positive. 3) Emphysema assessment included
several parts of soft tissue and bones also leading to false-positive results.
This led to 30 subjects with pulmonary emphysema. 30 of the remaining 157 participants
of our lung cancer screening program who did not show pulmonary emphysema were randomly
selected as control subjects.
MR images of the 30 subjects with pulmonary emphysema according to the LDCT definition
mentioned above (mean age: 60.3 ± 6.4 years) were presented to two radiologists with
6.5 years and 16 years of experience, respectively, together with the 30 control subjects
(mean age: 58.3 ± 5.8 years). The datasets were anonymized and presented in random
order. The MR images were evaluated in consensus. Again, the presence of emphysema
was first assessed qualitatively with predefined image windowing using the same three-point
scales as for LDCT. Then, automated quantitative analysis was performed using a software
program custom written in MATLAB (The MathWorks, Inc., Natick, Massachusetts, USA).
With the help of this program, the lungs were segmented from soft tissue, bones and
large vessels using a region growing algorithm with a threshold value of < 50 % of
the mean muscle signal obtained at 3 different positions across the imaging volume.
Voxels erroneously classified as lung tissue (e. g. trachea) were manually removed
from the lung segmentation. A second threshold value of < 15 % of the muscle signal
was used to define emphysematous lung parenchyma. Both thresholds followed the study
of Roach et al. on the performance of ultrashort echo time (UTE) MRI for evaluating
pulmonary emphysema [13 ], but were adjusted empirically to the T2-weighted sequence used in this present
study corresponding well to the – 930 HU selected for emphysema assessment on LDCT.
The chosen MRI thresholds were determined by one author of the manuscript who was
not taking part in qualitative and quantitative emphysema analysis to avoid a bias.
The thresholds were chosen after investigating other different thresholds (e. g. < 70 %
for lung segmentation and < 10 % for definition of emphysema).
Statistical analysis
Statistical analysis was performed with SPSS 24 (IBM, Armonk, New York, USA). Spearman
coefficient was applied for correlation of qualitative scores between MRI and LDCT
(presence/severity and morphological patterns). Pearson coefficient, linear regression
analysis and Bland-Altman plot were applied for comparison of emphysema indices as
calculated by MRI and LDCT. The Mann-Whitney U-Test was used to define differences
of qualitative scores and emphysema indices between MRI and LDCT.
Results
All 30 cases with pulmonary emphysema according to low-dose CT (LDCT) were accurately
detected by MRI. There were 3 subjects who seemed to have emphysema on MRI according
to qualitative assessment, yet they did not show emphysematous changes on LDCT (false-positive
results). One of these subjects showed multiple centrilobular lucencies and a slightly
elevated emphysema index (EI) of 5.4 % according to MRI, while the LDCT did not show
emphysema and an EI of 0.9 %. LDCT and corresponding MR images of this case are shown
in [Fig. 1 ]. In the other two cases, quantitative measurement on MRI did not correspond to emphysema
with indices of 2.7 % and 2.1 %, respectively.
Fig. 1 Qualitative and software-based quantitative assessment of pulmonary emphysema in
a subject without emphysema according to CT (a, c ; emphysema index of 0.9 %). This subject showed false-positive signs of emphysema
on MRI (b, d ; centrilobular lucencies and emphysema index of 5.4 %).Abb. 1 Qualitative und software-basierte quantitative Emphysemanalyse bei einem Patienten
ohne Emphysem in der CT (a, c ; Emphysemindex von 0,9 %). Dieser Patient zeigte falsch-positive Zeichen eines Emphysem
in der MRT (b, d ; zentrilobuläre Aufhellungen und Emphysemindex von 5,4 %).
The mean qualitative emphysema score was significantly higher in the emphysema group
compared to the control group for MRI (1.47 vs. 0.10, p < 0.001) and for LDCT (1.50
vs. 0.0, p < 0.001). The scores regarding severity and morphological patterns of emphysema
correlated significantly between MRI and LDCT (r = 0.912 and p < 0.001 for severity
in the emphysema group; r = 0.668 and p < 0.001 for severity in the control group;
r = 0.843 and p < 0.001 for emphysema pattern in the emphysema group; r = 1000 and
p < 0.001 for emphysema pattern in the control group). [Fig. 2 ] shows an example of the qualitative assessment of emphysema.
Fig. 2 Qualitative assessment of pulmonary emphysema in a subject with centrilobular and
paraseptal bullae on CT a and MRI b . The slight discrepancies between the images are due to different breathing positions
(CT images were acquired in inspiratory breath-hold while acquisition of MR images
was gated to the expiratory phase of the respiratory cycle).Abb. 2 Qualitative Bewertung eines Lungenemphysems bei einem Patienten mit zentrilobulären
und paraseptalen Bullae in der CT a und in der MRT b . Die geringen Unterschiede beruhen darauf, dass die CT-Aufnahmen in Inspiration aufgenommen
wurden, während die Akquisition der MRT-Aufnahmen auf die Phase der Exspiration getriggert
wurde.
[Table 1 ] shows the number of morphological emphysema patterns as detected by MRI and LDCT.
There were 3 cases in the emphysema group, for which MRI assigned a different emphysema
pattern than LDCT (centrilobular instead of mixed pattern, paraseptal instead of mixed
pattern, and mixed instead of centrilobular pattern in one case each).
Table 1
Number of different morphological patterns of emphysema as qualitatively assigned
by MRI and CT.Tab. 1 Anzahl an verschiedenen Emphysemformen wie sie in der MRT und in der CT vergeben
wurden.
MRI
CT
centrilobular
13
13
paraseptal
1
0
mixed or advanced destructive
16
17
The semi-automated software-based segmentation of the lung from the surrounding soft
tissue, bones and vessels on MRI was technically successful in all 60 cases. The manual
effort for the correction of lung segmentation was less than 3 minutes per case. Representative
images of quantitative emphysema analysis are shown in [Fig. 3 ].
Fig. 3 Software-based quantitative emphysema analysis in a subject of the emphysema group
with an emphysema index of 35.5 % on CT a and 32.5 % on MRI b , and in a control subject with an emphysema index of 0.4 % on CT c and 0.5 % on MRI d .Abb. 3 Software-basierte quantitative Emphysemanalyse bei einem Patienten aus der Emphysemgruppe
mit einem Emphysemindex von 35,5 % in der CT a und 32,5 % in der MRT b , und bei einer Kontrollperson mit einem Emphysemindex von 0,4 % in der CT c und 0,5 % in der MRT d .
The emphysema index was significantly higher in the emphysema group for MRI and LDCT
(p < 0.001) ([Table 2 ]) with significant correlation between MRI and LDCT (r = 0.960 and p < 0.001 for
emphysema group; r = 0.746 and p < 0.001 for control group). The Bland-Altman plot
and linear regression analysis are shown in [Fig. 4 ].
Table 2
Mean emphysema index (EI) as calculated on MR and CT images.Tab. 2 Durchschnittlicher Emphysem-Index (EI) berechnet in MRT- und CT-Aufnahmen.
EI MRI
EI LDCT
Emphysema group
11.6 ± 11.3
11.6 ± 10.3
Control group
1.4 ± 1.2
0.7 ± 0.7
Fig. 4 Linear regression of emphysema indices (EI) for emphysema group a and control group b , as well as Bland-Altman plot of emphysema indices for emphysema group c and control group d .Abb. 4 Lineare Regression der Emphysem-Indices (EI) für die Emphysemgruppe a und die Kontrollgruppe b , sowie Bland-Altman-Diagramm der Emphysem-Indices für die Emphysemgruppe c und die Kontrollgruppe d .
Discussion
The main finding of this study is that T2-weighted PROPELLER MRI, which has been shown
to be suitable for lung imaging [18 ]
[19 ]
[20 ]
[21 ], may also be used for the assessment of pulmonary emphysema despite its lower spatial
resolution compared to CT. The results are comparable to previously published studies
using UTE imaging and functional MRI in subjects with emphysema [10 ]
[11 ]
[12 ]
[13 ]
[15 ], while being technically more easy to implement.
The software-based quantitative assessment of pulmonary emphysema on CT images is
well established. Regarding normal-dose CT with a slice thickness of 1 mm, a threshold
of – 950 HU seems to be optimal for CT densitometry analysis of emphysema [4 ]
[5 ]
[6 ]
[22 ]. Yet, in a statement of the Fleischner Society, Lynch et al. pointed out that excessive
image noise with a reduced CT dose can simulate emphysema, particularly on quantitative
CT [4 ]. Moreover, image quality and noise level with simulated mAs levels below 60 mAs
were significantly inferior to images with higher simulated mAs levels in a study
by Ley-Zaporozhan et al. [23 ]. They concluded that imaging dose could be lowered to 60 mAs in thin-slice CT without
a diagnostically relevant increase in noise impairing image quality. Hence, the optimal
threshold for emphysema quantification with low-dose CT using mAs levels below 30 mAs
as in our study has yet to be determined. After testing different thresholds from
– 910 to – 950 HU, we decided to use a midway threshold of – 930 HU in our study,
since this value correlated best with the results of the qualitative emphysema analysis
(in many cases with a threshold of – 950 HU; lucencies that were apparently related
to emphysema were not indicated as such by the CT software tool, while a threshold
of – 910 HU led to an obvious overestimation in visual analysis). This is in contrast
with a study of Gierada et al., who showed that there were no significant differences
between normal-dose and LDCT for emphysema analysis [24 ]. However, the reconstructed slice thickness in their study was 5 mm and the tube
current-exposure time product was 30 – 60 mAs, while slice thickness and radiation
dose of LDCT were much lower in our study (2 mm and 25 mAs). In addition, the absolute
CT threshold was not particularly important for this study, since our intention was
the correlation with MRI, and not to find the most suitable threshold for densitometry
analysis on LDCT.
Based on our study results, pulmonary emphysema may be assessed quantitatively with
structural T2-weighted PROPELLER MRI. Still, it should be mentioned that the correlation
was excellent when emphysema was present, but a little less valid in subjects without
emphysema.
To the best of our knowledge, there is only one study by Roach et al. [13 ] including software-based emphysema quantification using structural MRI (UTE). However,
high-resolution UTE imaging of the lung might not be feasible in the clinical routine,
since it is technically demanding and might take up to 20 minutes for image acquisition
[12 ]
[13 ]. In comparison, most clinical MRI scanners should be able to yield a T2-weighted
PROPELLER sequence similar to the one being used in our study. Thus, the presented
approach may be more transferable to clinical routine. Our study results suggest that
T2-weighted PROPELLER MRI may have the potential to be used for the quantification
and phenotyping of severe pulmonary emphysema, and subsequently for the identification
of progression in follow-up examinations to avoid the radiation exposure of repeated
CT scans. At the same time, this approach would allow for the detection of other relevant
findings, such as pulmonary nodules or inflammatory changes as previously shown [10 ]
[12 ]
[21 ].
Our study has several limitations. First, this is a retrospective study, since the
images were not primarily acquired for the analysis of pulmonary emphysema. However,
the intention of the present study was to evaluate whether emphysema could be assessed
with a conventional T2-weighted PROPELLER sequence that can be easily applied in the
clinical routine. Second, the results of emphysema analysis were not correlated with
spirometric parameters such as the forced expiratory volume in the first second-forced
vital capacity-ratio (FEV1/FVC). This is desirable in future studies. The third limitation
is the low sample size of our study population. A fourth limitation is that some LDCT
datasets with distinct noise overlay (especially seen in adipose individuals) or extensive
atelectasis/infiltrates were excluded from further analysis with MRI, which may have
led to a preselection bias. And fifth, subjects with mild emphysema were not part
of the analysis. Even though it was not evaluated in this study, we have to assume
that the current T2-weighted PROPELLER sequence would probably not be capable of detecting
slight emphysematous changes as reliably as severe emphysema due to the much lower
spatial resolution compared to LDCT. Still, this did not have a major influence on
the visual scoring.
In conclusion, the presence and extent of pulmonary emphysema may be assessed qualitatively
and quantitatively using T2-weighted PROPELLER MRI with very good correlation to LDCT
according to the present study. However, our study results should be validated in
larger prospective studies.
