Key words
computed tomography - magnetic resonance imaging - segmentation - muscle atrophy -
sarcopenia - swallowing muscles
Purpose
Different approaches for the morphometric assessment of skeletal muscles using a variety
of imaging techniques have been reported. In contrast, changes in morphometry of the
head and neck muscles have received little attention so far. Only initial results
using two dimensional approaches in computed tomography (CT) [1] and comparing magnetic resonance imaging (MRI) and ultrasonographic measurements
[2] have been published.
Several suprahyoid muscles play a key role in the characteristic movement of the hyoid
bone during swallowing. For example, the geniohyoid (GH) muscle connects the posterior
aspect of the mandible in the midline with the anterior surface of the body of the
hyoid bone. The GH muscle’s contraction drives the hyoid bone upward and forward together
with the mylohyoid, stylohyoid, and anterior belly of the digastric muscles [3]. Sarcopenia of these muscles may play an important role in reducing hyoid bone movement
and thus results in an increased risk of dysphagia in affected patients [1].
Feng et al. [1] found an interesting association between atrophy of the geniohyoid muscle and aspiration
status in older adults using simplified CT measurements. In a recent publication our
group described an association of age and severity of dysphagia with muscle atrophy
using three-dimensional semi-automated segmentation of different swallowing muscles
[4]. However, ionizing radiation poses an ethical barrier to research when using CT
imaging to investigate muscle size in prospective non-patient populations. So far,
no studies have compared CT and MRI regarding assessment of this muscle group. Therefore,
there is a need to prove the efficacy of techniques that rely on non-radiating methods
and produce high-resolution soft tissue images, such as MRI.
This study aims to compare the muscle volumes of submental muscles (geniohyoid muscle
and anterior belly of the digastric muscles) obtained from both CT and MRI images
to see whether equal muscle volumes can be achieved with magnetic resonance imaging.
If MRI provided similar results in the segmentation of swallowing muscles, changes
in the musculature responsible for swallowing could be examined without ionizing radiation
in prospective studies.
Methods
Study design and setting
A retrospective observational study was conducted using data of patients who had undergone
computed tomography angiography and magnetic resonance imaging of the head and neck
region at the Department of Clinical Radiology at the University Hospital of Muenster
within a time frame of less than 50 days. All patients had undergone CT first and
MRI afterwards using contrast agent because a neoplasia in the head and neck region
was suspected.
Patients were included if: 1) they had undergone CT and MRI within a time frame of
less than 50 days and 2) a neoplasia in the head and neck region was suspected. No
patients were excluded.
Of the 21 included patients, 11 (52.4 %) were female and 10 (47.6 %) were male. The
median age was 46.6 (35.6 – 57.8) years.
All examinations were part of routine clinical care. The study was approved by the
local ethics committee. As the nature of the study was purely retrospective, the review
board waived the need for informed consent. The identity of each documented patient
is completely anonymized.
Imaging protocol and data post-processing
CT scans were performed on a 128-slice dual-source CT scanner (Somatom Definition
Flash; Siemens Medical Solutions, Forchheim, Germany). CTA was obtained with the following
parameters: 120 kV, 175 mAs, 1.0-mm slice reconstruction, 1-mm increment, 0.6-mm collimation,
0.8 pitch, H20f soft kernel. In all scans contrast medium was used (80 mL Ultravist
370 and 50 mL NaCl flush at 4 mL/s, scan start 6 seconds after bolus tracking at the
level of the ascending aorta).
MRI was performed using a 3 Tesla MRI (Philips Ingenia 3.0 T, Philips, Eindhoven,
The Netherlands). We obtained coronal and axial T1-weighted images (TE/TR = 7508/595 267 ms,
TI = 0 ms, slice thickness = 4.4 mm) using a phased array coil. In all patients 1 mmol/kg
of gadolinium-based contrast agent (Gadolinium Omniscan; GE Healthcare, Chicago, IL,
USA) was administered by the technical assistant.
On coronal CTA and MRI T1-weighted images, the geniohyoid muscle and the anterior
parts of the digastric muscles were segmented slice by slice on both sides using Medical
Imaging Toolkit (MITK 2.4.0.0, Heidelberg, Germany). In the following steps segmentations
were adjusted semi-automatically in the other dimensions ([Fig. 1], [2]). This was done by two different readers blinded to the patient’s swallowing function
and all clinical information.
Fig. 1 CT segmentation process of the anterior body of the digastric muscle (left and right)
and the geniohyoid muscle. A–E First the muscle was segmented slice by slice in coronal CT images (here exemplary
the left digastric muscle). In the next step three-dimensional muscle volumes were
calculated automatically using the programs interpolation function. F, G Finally the accuracy of the coronal segmentation was checked using the axial images
and the sagittal multiplanar reconstructions. If necessary, corrections of the segmentation
were performed using all three dimensions. Image H shows the 3 D visualization of the left digastric muscle as segmented before.
Abb. 1 Illustration der CT Volumetrie des Musculus digastricus und Musculus geniohyoideus.
A–E Zunächst wurde der Muskel Schicht für Schicht in den coronalen CT-Bildern segmentiert
(hier exemplarisch der linke Musculus digastricus). Im nächsten Schritt werden automatisch
mittels Interpolationsfunktion des Programms 3-dimensionale Muskelvolumina berechnet.
F, G Abschließend wird die Richtigkeit der coronalen Segmentierung in den axialen Schichten
und den sagittalen Rekonstruktionen überprüft und gegebenenfalls korrigiert. H zeigt die 3D-Visualisierung des linken Musculus digastricus nach der Segmentation.
Fig. 2 MRI segmentation process of the anterior body of the digastric muscle (left and right)
and the geniohyoid muscle. A–C Analog to CT, the muscle was segmented slice by slice in coronal MR images (here
exemplary the left digastric muscle). D, E In the next step the accuracy of the coronal segmentation was checked using the axial
images and if necessary corrections of the segmentation were performed.
Abb. 2 Illustration der MRT Volumetrie des Musculus digastricus und Musculus geniohyoideus.
A–C Analog zum CT wurde der Muskel in den coronalen Sequenzen Schicht für Schicht volumetriert
(hier exemplarisch der linke Musculus digastricus). D, E Im nächsten Schritt wurde die Genauigkeit der Segmentation in den axialen Schichten
überprüft und anschließend gegebenenfalls Änderungen vorgenommen.
Statistical analysis
Univariable distribution of metric variables is described by median and interquartile
range. Differences between CT and MRI muscle volumes were tested using a test of equivalence.
We defined an equivalence range of x = 200 mm3. The two diagnostic modalities were defined as equivalent when the differences between
CT and MRI volumes were less than x.
Accordingly, for each of the three muscles (i = 1/2: left/right anterior belly of
the digastric muscle, i = 3: geniohyoid muscle), the following one-sided null hypotheses
were tested by Wilcoxon signed-rank test on a multiple significance level of 5 %:
“CT volume ≥ MRI volume + x” (Hi+) and “CT volume ≤ MRI volume – x” (Hi-). If null hypotheses Hi+ and Hi- can both be rejected for some i, the level of deviation between both modalities is
less than x for muscle i (|CT.vol – MRI.vol | < x) and consequently both modalities
have been shown to be equivalent for muscle i.
Adjustment for multiple testing was performed using the closed testing procedure [5] with each intersection null hypothesis being tested using a Bonferroni test while
taking into account that the intersection of null hypotheses Hi+ and Hi- is empty for each i = 1, 2, 3.
In order to investigate a potential time effect on the muscle volumes, we calculated
muscle volume differences between MRI and CT (∆ "MR-CT" volume in mm3) for each muscle and assessed the correlation with the time interval between the
modalities using Spearman’s rho. Interrater reliability of the segmented muscle volumes
was quantified using the intra-class correlation coefficient (ICC).
Statistical analyses were performed in SPSS version 24 (IBM Corporation, Armonk NY).
Results
All included patients had tumors of the head and neck region (14 laryngeal cancers,
6 pharyngeal cancers, 1 sarcoma).
The median volumes for each anterior belly of the digastric muscle derived from CT
were 3051 mm3 (left) and 2969 mm3 (right), and those derived from MRI were 3218 mm3 (left) and 3027 mm3 (right). The median volume of the geniohyoid muscle was 6580 mm3 on CT and 6648 mm3 on MRI ([Table 1]). Interrater reliability was high for all segmented muscles; values on CT were 0.996
(95 % confidence interval = 0.991; 0.999) for the left anterior belly of the digastric
muscle, 0.994 (0.986; 0.998) for the right anterior belly of the digastric muscle
and 0.998 (0.995; 0.999) for the geniohyoid muscle and for MRI they were 0.976 (0.942;
0.990); 0.998 (0.994; 0.999); 0.999 (0.998; 1.000), respectively.
Table 1
Comparison between CT and MRI muscle volumes of the anterior body of the digastric
muscles (left and right) and the geniohyoid muscle.
|
muscle volumes in mm3
|
median (IQR)
|
minimum
|
maximum
|
|
CT left anterior body of the digastric muscle, mm3
|
3051 (2808; 3612)
|
912
|
4901
|
|
MRI left anterior body of the digastric muscle, mm3
|
3092 (2555; 3463)
|
852
|
5552
|
|
CT right anterior body of the digastric muscle, mm3
|
2969 (2787; 3618)
|
1412
|
5003
|
|
MRI right anterior body of the digastric muscle, mm3
|
3027 (2665; 3580)
|
1305
|
5101
|
|
CT geniohyoid muscle, mm3
|
6580 (5891; 8145)
|
2531
|
9012
|
|
MRI geniohyoid muscle, mm3
|
6648 (5900; 8145)
|
2474
|
8985
|
CT = computed tomography; MRI = magnetic resonance imaging; IQR = interquartile range.
All six null hypotheses regarding CT- and MRI-based segmentation could be rejected
at a multiple significance level of 5 % ([Table 2]).
Table 2
Equivalence test of CT and MRI muscle volumes of the anterior body of the digastric
muscles (left and right) and the geniohyoid muscle.
|
muscle volume
|
CT volume, median (IQR)
|
MRI volume, median (IQR)
|
p-value of Hi+/Hi–
|
|
1
|
left anterior body of the digastric muscle, mm3
|
3051 (2808; 3612)
|
3092 (2555; 3463)
|
0.001/0.001
|
|
2
|
right anterior body of the digastric muscle, mm3
|
2969 (2787; 3618)
|
3027 (2665; 3580)
|
< 0.001/< 0.001
|
|
3
|
geniohyoid muscle, mm3
|
6580 (5891; 8145)
|
6648 (5900; 8145)
|
< 0.001/< 0.001
|
For each of the three muscles (i = 1, 2, 3) the following one-sided null hypotheses
were tested by Wilcoxon signed-rank test: “CT volume ≥ MRT volume + x” (Hi+) and “CT volume ≤ MRT volume – x” (Hi–). We defined an equivalence range of x = 200 mm3. Given are multiplicity adjusted p-values for each null hypothesis.
The mean time interval between the CT and MRI examinations was 34 days (IQR 25; 41).
The muscle differences of each muscle between the two modalities did not reveal a
noticeable correlation to the time interval between the examinations (“MRI – CT” volume:
left and right anterior bellies of the digastric muscle (r = 0.003 and r = –0.008)
and the geniohyoid muscle (r = 0.075).
Discussion
This is the first study to correlate the measurement of swallowing muscles using CT
and MRI images. We included the anterior part of the digastric muscles on both sides
and the geniohyoid muscle for their good contrast and discrimination in both techniques.
Our results show that volume segmentation in both techniques is equal within an equivalence
range of x = 200 mm3. Moreover, the high interrater reliability suggests a good reproducibility of our
results in future studies.
So far, automated muscle segmentation has been used for example to quantify skeletal
muscles of the abdomen and of the paravertebral lumbar muscles on CT and MR images
[6]
[7]. The technique we used has been established for more than a decade for different
segmentation processes [8] including adipose tissue measurements [9] and segmentation of bone metastases [10]. In the clinical routine muscle segmentation can be useful especially when deployed
automatically to determine whole-body muscle mass and thereby quantify the progress
of chronic diseases or conditions, i. e., (autoimmune) myositis or sarcopenia [6]
[7].
With regards to the swallowing musculature, previous studies used less precise CT-based
two-dimensional parameters, in particular diameters of the geniohyoid muscle [1], CT-based evaluations of the density of the masseter and the medial pterygoid [11] and tongue thickness measurement with ultrasound [12].
Advantages of magnetic resonance imaging include better soft-tissue contrast compared
with CT images allowing for better discrimination of above-mentioned muscles even
for inexperienced raters. CT and MRI images are both easily reproducible offering
an objective follow-up for each patient. According to our results, the potential disadvantage
of MRI with longer acquisition times potentially leading to more swallowing artifacts
does not seem to be relevant for muscle segmentation processes. Although the coil
used for MRI dictates the head position to some extent with the possible consequence
of a decreased length of the submental muscle group, we did not observe a systematical bias
when deploying a multi-dimensional segmentation approach.
Our findings have implications for swallowing research and clinical management of
dysphagia. A previous study of our group that found an association between atrophy
of the submental muscles, age and severity of dysphagia was based on CT [4]. However, prospective studies require noninvasive assessment without ionizing radiation.
Quantifying decreases in submental muscle volumes may provide insight into the underlying
mechanisms of dysphagia caused by sarcopenia (the degeneration of skeletal muscle
associated with aging) or muscle weakness.
Some limitations of this study that are due to the retrospective nature of data acquisition
need to be addressed. We chose coronal and axial T1- weighted images as the MRI images.
Even though our results confirm the equality of CT and MRI segmentation, three-dimensional
MR images would probably allow for even more accurate measurements and should therefore
be deployed in future prospective studies. The period in between the CT and MRI scans
may have presented another possible limitation. To minimize the chance of a relevant
muscle gain or loss during this lag time, we included only patients with a relatively
short time span of less than 50 days between the two examinations (median follow-up
time in our study was 34 days). Therefore, in our study the time between the two examinations
showed no relevant impact on the muscle volume differences of both methods.
Conclusion/clinical relevance
CT-based segmentation and MRI-based segmentation of the digastric and geniohyoid muscle
are equally feasible. The potential advantage of MRI for prospective studies is the
absence of ionizing radiation.
