Keywords
acromion - magnetic resonance imaging - shoulder impingement syndrome - radiography
- rotator cuff tear
Introduction
The etiology of rotator cuff tears (RCT) is still uncertain and it's now believed
that it is multifactorial.[1] Factors that may contribute to the occurrence of those tears can be divided into
intrinsic or extrinsic. Intrinsic factors include age,[2] tendinous degeneration,[3] genetic aspects,[4]
[5]
[6] smoking,[7]
[8] diabetes,[9] alcohol abuse.[10] Historically, extrinsic factors are those related to the impingement between acromion
process and the rotator cuff, specifically supraspinatus tendon.[1] Since Neer postulated that 95% of RCT were caused by acromial impingement, the influence
of scapular morphology in the etiology of those tears has been exhaustively investigated.[11] Following the same reasoning, Bigliani observed that an anterior and inferior acromial
inclination could lead to supraspinatus tears.[12] However, subsequent studies contested this finding and suggested that tendon degeneration
precedes acromial spur formation, leading to dynamic humeral head superior migration
and therefore to secondary acromial impingement.[1]
In 2006, Nyffeler et al.[13] suggested that a large lateral (not anteroinferior) extension of the acromion related
to a higher incidence of RCT. Authors postulated that a larger lateral extension of
the acromion predisposes to supraspinatus degeneration by means of increasing deltoid
shear forces and leading to superior migration of humeral head, consequently causing
supraspinatus impingement against the acromion.[13] Authors recommended then that the lateral extension of the acromion should be measured
through the acromial index (AI), which is the relation between two distances: from
glenoid surface to the lateral acromion extremity and from glenoid surface to the
lateral humeral cortex.[13] However, AI may be influenced by humeral anatomy, such as in deformity and malunion
cases. To solve this shortcoming, Moor et al.[14] developed the critical shoulder angle (CSA), which depends only on scapular anatomy.
This angle is formed between a line running from the superior to the inferior pole
of the glenoid and another one running from the latter to the lateral acromial extremity.
Authors found that 84% of their patients with rotator cuff tears had a CSA higher
than 35°. Afterwards, the relationship between a high CSA and RCT has been suggested
by several authors,[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25] as well as the higher retear risk after surgical treatment of those tears.[26]
[27]
[28]
[29]
[30]
Both AI and CSA are measured in true AP view of the shoulder joint and the scapular
positioning is critical to the reproducibility of radiographic parameters.[13]
[14] Positioning errors may lead to inconsistence and heterogeneity of AI and CSA measurements.[24]
[31]
[32]
[33] Currently, the gold standard imaging modality in the painful shoulder is magnetic
resonance image (MRI), due to its high sensibility, specificity and accuracy in diagnosing
RCT.[34] Once standardizing images acquisition is easier and more reliable in MRI than in
radiographs,[34] one may infer that measurements of AI and CSA in MRI are more accurate than in radiographs.
However, results of papers on this subject are conflicting.[35]
[36]
Therefore, the primary objective of this study is to compare the interobserver and
intra-observer agreement for AI and CSA values measured in both radiographs and MRI
of the shoulder. Also, we aim to compare absolute values of AI and CSA obtained in
these image modalities, assessing whether MRI is a reliable method in determining
both anatomical parameters.
Materials and Methods
Work approved by the ethics committee of our institution (document number: 32689114.7.0000.5257).
Study Design and Subjects Selection
This is a blind prospective longitudinal observational study. Skeletally mature patients
who had medical indication of investigating shoulders conditions through radiographs
and MRI were included. The exclusion criteria were previous history of shoulder fracture
or surgery and those whose image exams revealed humeral or scapular bony deformity.
Imaging
After giving their written consent, patients were referred to radiology department
to take both radiographs and MRI in the same day. Radiographs were taken in standing
position with the shoulder in neutral rotation. The proper positioning of the patient
to obtain a true AP view was made under fluoroscopic control (Axiom Iconos MD; Siemens,
Erlangen, Germany). Only A1 images in the Suter-Henninger system[31] were accepted and every radiograph out of this standard was repeated. MRI exams
were performed in high filed, closed machines, with a 1,5 T magnet (Magnetom Avanto;
Siemens, Erlangen, Germany). The researchers studied T2-wheighted images with fat
suppression in the axial, coronal and sagittal planes; T1 and T2-wheighted images
without fat suppression in the axial, coronal and sagittal planes.
Images Analysis
Images of both radiographs and MRI exams were recorded in portable media and taken
to two examiners, both fellowship trained shoulder surgeons, with different levels
of experience (one and 15 years). Images were imported to a DICOM viewer (Radiant
DICOM Viewer, Medixant, Poznan, Poland) and analysis of the radiographs were made
according to Moor et al.[14] and Nyffeler et al.[13] ([Figs. 1] and [2]); measurements in MRI were made according to Spiegl et al.[35] description ([Fig. 3]). The evaluators had access to the complete examination, with the full sets of images.
Fig. 1 True AP view radiograph of the shoulder, showing CSA measurement. The angle is formed
between two lines: one from the superior to the inferior pole of the glenoid, and
other from the latter to the lateral edge of the acromion.
Fig. 2 True AP view radiograph of the shoulder, showing anatomical parameters required to
measurement of the AI. This index is obtained dividing the distance from the glenoid
surface to the most lateral edge of the acromion (GA) by the distance from the glenoid
surface to the lateral cortex of the proximal humerus (GH). AI = GA/GH.
Fig. 3 Frequently, in MRI the most lateral edge of the acromion is not in the same plane
of the glenoid midline. Thus, we used the cursor to mark the lateral acromion (3A)
and then scroll the images until the glenoid midline (3B), where the measurements
are made.
None of the examiners had access to the names of the patients and only the main searcher
knew the identity of subjects. Both evaluators and patients received coded numbers
to identify them. Radiographs and MRI were given separate numbers so that evaluators
could not relate the exams of a same patient. Twelve weeks after the first evaluation,
exams were once more presented to examiners and a second evaluation was conducted.
Statistical Analysis
Statistical analysis was performed using GraphPad Prism version 9.0.0 (121) for Windows
64-bit (GraphPad Software, LLC), Stata/MP 16.1 for Windows (64-bit x86–64–StataCorp,
LLC), and StatMate 2 for Windows (GraphPad Software, LLC). Continuous variables were
given as a mean ± standard deviation. The normality distribution of the continuous
variables was tested by the Kolmogorov-Smirnov test. Inter- and intra-observer reliability
was presented as an Intraclass Correlation Coefficient (ICC) and agreement was classified
according to Landis and Koch[37] criteria: a value inferior to 0.01 describes a poor agreement; a value between 0.01
and 0.20 describes a slight agreement; 0.21 to 0.40, a fair; 0.41 to 0.60, a moderate;
0.61 to 0.80, a substantial; and 0.81 to 1.00, an almost perfect agreement. The differences
between two measurements were evaluated using Bland-Altman plots. The study had a
90% power to detect a smallest average difference between pairs of 0.09 in the CSA
and 0.002 in the IA results with a significance level (α) of 0.05 (two-tailed). The
significance level for all tests was set at p < 0.05.
Results
Demographics and general characteristics of the sample are depicted in [Table 1]. We evaluated 134 shoulders in 124 subjects, with a mean age of 52 years old (ranging
18 to 85); there were 68 females and 56 males. Dominant side was affected in 89 subjects
and 10 patients had bilateral complaints. Isolated pain was the main complain in 116
shoulders and pain associated to stiffness were reported in 8 shoulders. Isolated
weakness was seen in one shoulder and instability was the main complaint in nine shoulders.
The mean length of symptoms was 23 months (ranging 0,1 to 400). Full thickness and
partial thickness RCT were found in 36 and 33 patients, respectively. Supraspinatus
tendon was the most committed one, followed by infraspinatus and subscapularis; there
was no teres minor tendon tear in this sample ([Table 2]).
Table 1
|
124 subjects - 134 shoulders
|
|
|
Mean age (years)
|
52 (18–85)
|
|
Gender (n)
|
|
|
Female
|
68
|
|
Male
|
56
|
|
Affected side (n)
|
|
|
Dominant
|
89
|
|
Bilateral
|
10
|
|
Main complaint
|
|
|
Pain
|
116
|
|
Instability
|
9
|
|
Stiffness
|
8
|
|
Weakness
|
1
|
|
Duration of symptoms (months)
|
23,4 (0,1–400)
|
Table 2
|
Normal
|
Tendinosis
|
Partial tear
|
Full thickness tear
|
|
Supraspinatus (n)
|
34
|
33
|
31
|
36
|
|
Infraspinatus (n)
|
79
|
39
|
4
|
12
|
|
Subscapularis (n)
|
107
|
19
|
5
|
3
|
High ICC values were observed for intra-observer reliability, regarding both CSA and
IA measured either in MRI or radiographs ([Table 3]). Therefore, there was an excellent, almost perfect intra-observer agreement for
both CSA and AI measurements made in MRI and radiographs. There was an almost perfect
interobserver agreement for both CSA and AI measured in radiographs and a substantial
interobserver agreement for measurements made in MRI ([Table 4]). Absolute values found for AI and CSA were also correlated in both image modalities
used in this study. ICC values for AI and CSA found in radiographs and MRI for both
observers were 0.86 and 0.87, respectively. Bland-Altman plots show high inter-method
correlation for both observers regarding either radiographs and MRI ([Fig. 4]). Mean difference for AI values measured in X-rays and in MRI was 0.01 and 0.03
for observers 1 and 2, respectively. Mean difference for CSA values obtained in X-rays
and MRI was 0.16 and 0.58 for observers 1 and 2, respectively.
Fig. 4 Bland-Altman plots showing difference vs. average distribution of AI (A, B) and CSA (C, D) indexes measured in X-rays and in MRI for observers 1 and 2, respectively.
Table 3
|
Interclass correlation coefficient
|
|
Observer 1
|
Observer 2
|
Mean
|
|
CSA
|
|
|
|
|
X-ray
|
0.979*
|
0.893*
|
0.936
|
|
MRI
|
0.975*
|
0.905*
|
0.940
|
|
AI
|
|
|
|
|
X-ray
|
0.916*
|
0.900*
|
0.908
|
|
MRI
|
0.929*
|
0.915*
|
0.922
|
Table 4
|
Interclass correlation coefficient
|
|
T1
|
T2
|
Mean
|
|
CSA
|
|
|
|
|
X-ray
|
0.910*
|
0.875*
|
0.892
|
|
MRI
|
0.737*
|
0.768*
|
0.752
|
|
AI
|
|
|
|
|
X-ray
|
0.836*
|
0.863*
|
0.849
|
|
MRI
|
0.737*
|
0.634*
|
0.685
|
Discussion
This study found high inter and intra-observer agreement for AI and CSA measured in
both radiographs and MRI exams. ICC values for intra-observer agreement were even
higher than those for inter-observer agreement, reflecting that observers tend to
agree more with themselves than with each other. Although quite similar, intra-observer
agreement was slightly higher in MRI than in radiographs, for both AI and CSA. However,
inter-observer agreement was higher in radiographs than in MRI exams; even so, ICC
values for inter-observer agreement in MRI were still considered high and a substantial
agreement was found. Not only observers agreed with themselves and with each other,
we also had a high inter-method correlation – absolute AI and CSA values observed
in radiographs and in MRI were very similar and high ICC values were observed on this
analysis. These findings may suggest that either MRI and radiographs are equally suitable
for measurements of both AI and CSA.
In fact, MRI has long become the main diagnostic tool in investigating shoulder pain,[34] due to its high accuracy in detecting ligamentous, tendinous and bony injuries.
Besides that, the acquisition of the proper scapular plane is easier in MRI than in
routine radiographs, since it's done by the radiology technician immediately before
the exam begins. Fortunately, the radiography system we used in this study allowed
for adequate patient positioning under fluoroscopic control, assuring a true AP view
of the shoulder. Nonetheless, this may not be available for routine use in most of
the orthopedics services around the world. Also, one must note that even when using
standard protocols and fluoroscopic control, obtaining a true AP view can be complicated
by many individual factors, such as medical comorbidities, variations in scapular
version and shape, age, body habitus, etc. True AP views might be identified by ruling
out exams showing double contoured glenoids and also those exams showing flexion or
extension malpositioning of the scapula, which is assessed by coracoid position regarding
its overlap with glenoid. In this way, we found Sutter-Henninger classification[31] useful to exclude radiographs made with malpositioned scapula. Authors noted that
when doing so, 89% of CSA measurements were within less than 2° of accuracy. Even
respecting a standard radiography protocol, Chalmers et al.[38] retrospectively observed that only 19% of radiographs in their study were suitable
to measure CSA, according to Sutter-Henninger classification. However, authors did
not used fluoroscopic positioning of the patient. As our study had a prospective design,
we could guarantee that only X-rays defined as A1 in Sutter-Henninger classification
were included.
When measuring both CSA and AI in MRI, one must consider that acromial most lateral
edge is not at the same plane of glenoid surface and it's generally slightly posterior
to it.[39] This is even more concerning for AI, which relies also on the localization of lateral
humeral cortex besides the glenoid surface and lateral acromial edge, i.e., there
are three anatomic variables instead of two. To overcome this, we used a simple, previously
described technique[35]
[36]: first, the most lateral part of the acromion was identified and marked with a cursor;
then the MRI slice which passes through the glenoid midline was selected and the measurements
were made. Although CSA and AI depend on the same anatomic references regardless the
diagnostic method used, one could expect disparate values measured in X-rays and in
MRI due to inherent differences between each of these imaging modalities. And even
we have observed high agreement values for both imaging modalities separated, this
could not necessarily mean that values found in radiographs were similar to those
found in MRI. For this reason, we used Bland-Altman plots to compare those values
and found that mean values for both AI and CSA obtained either in MRI or in radiographs
were almost identical. This finding may support the clinical use of MRI in measuring
AI and CSA as well it's use in future studies.
Our findings are in contrast with those reported by Spiegl et al.[35] They found high interobserver and intra-observer agreement for CSA measurements
made in X-rays, but lower agreement (moderate for interobserver and poor for intra-observer)
for measurements made in MRI. Curiously, authors also found a significant difference
in mean CSA values measured in radiographs versus MRI, only in osteoarthritis patients,
but not in those with RCT. They speculate that this discrepancy may be due to the
difficulty of defining glenoid borders in osteoarthritis patients. Although our sample
is much bigger, we had fewer patients with osteoarthritis in this series (seven versus
ten in Spiegl et al. study) and didn't notice this difference. Besides having a smaller
sample, they didn't give details on radiographic technique used in their study, which
may be a potential reason for the dissimilarity between our results and theirs.
Conversely, Incesoy et al.[36] measured CSA and AI in 870 subjects and found high inter- and intra-observer agreement.
They also reported that both AI and CSA were significantly related to full-thickness
RCT. Although authors stated that patients had also radiographs, only MRI data were
included in their paper; thus, a comparison between absolute AI and CSA values in
X-rays and in MRI was not made. Recently, Garcia et al.,[27] in a rather small series, found similar values in CSA measured both in radiographs
and in MRI. In their prospective, randomized, blind study, they also observed more
experienced evaluators to achieve higher agreement between those imaging modalities.
Our study has some strengths. First, we could use a solid standardized method for
radiographic exams, in which patients were positioned under fluoroscopic control.
As a prospective study, we could repeat every radiograph that didn't meet the criteria
for a true AP view of the shoulder. Also, both evaluators were fellowship-trained
shoulder surgeons, which may have contributed to the high agreement values obtained.
Besides, we had a high number of exams, which allowed for powerful statistical analysis.
By the other hand, this study also had some weaknesses. Although it's advisable that
strict true AP views of the shoulder should be used when investigating shoulder pain,
we acknowledge that this might be difficult in some patients and under certain conditions.
Therefore, the findings of our study may not be applicable to less than perfect true
AP radiographs. Also, the main indication for MRI in our series was to investigate
shoulder pain, mostly caused by rotator cuff tears. Roughly, two-thirds of our patients
had partial and full-thickness rotator cuff tears and we had few patients with other
diagnosis, such as instability, frozen shoulder, and osteoarthritis. So, our results
may not be reproductible in cases other than rotator cuff tendinopathy.
Conclusion
Both MRI and X-rays provided high intra- and interobserver agreement for measurement
of AI and CSA. Absolute values found for AI and CSA were highly correlated in both
image modalities. These findings suggest that MRI is a suitable method to measure
AI and CSA.
