Quantitative Diagnosis of Rotator Cuff Tears Based On Sonographic Pattern Recognition
Quantitative Diagnosis of Rotator Cuff Tears Based On Sonographic Pattern Recognition
The accurate diagnosis of rotator cuff disorders is important to determine treatment strat-
egy, especially differentiating tears from other types of tendinopathies [7, 8]. The presence of
rotator cuff full-thickness tears influence the decision to undertake the surgical repair or
arthroplasty [9]. Furthermore, the measurement of size and location of rotator cuff tear is nec-
essary for pre-operative planning. Clinically, imaging examinations such as shoulder ultra-
sound, roentgenogram, magnetic resonance imaging (MRI), and magnetic resonance
arthrography (MRA) are necessary for assessing rotator cuff tears because physical examina-
tions and clinical symptoms may be unreliable [10, 11]. Compared to other imaging modali-
ties, shoulder ultrasound has the advantages of being inexpensive, conducted in real-time, and
convenient to operate. Nevertheless, using ultrasound is operator dependent which relies on
adequate training and experience in diagnosing rotator cuff tears. Especially, differentiating
partial and full-thickness tears from other tendinopathies is a challenge [12]. According to pre-
vious literature [13–19], the diagnostic sensitivity and specificity of shoulder ultrasound on
tear detection has a range of 46%-95% and 50%-95%, respectively. The variabilities are highly
correlated with the level of experience of the operator and the patterns of the rotator cuff tears
[20]. Experienced musculoskeletal radiologists or shoulder orthopedic surgeons possess a
higher accuracy than general radiologists and ultrasonographers in diagnosing rotator cuff
tears via shoulder ultrasound [20]. From this viewpoint, the inter-observer variability in diag-
nosing rotator cuff tears between operators with different professions or levels of experience is
substantial, and higher variability is demonstrated in the diagnosis of partial-thickness tears
[21, 22]. The introduction of quantitative and automated diagnostic procedures could poten-
tially reduce the impact of variability.
Computer-aided diagnosis systems provide an objective, quantitative assessment of lesion
type and grade [23–26]. After defining the lesion area with manual or semi-automatic segmen-
tation, quantitative features can be extracted and combined in an artificial intelligence classi-
fier. By considering a broad range of relevant features, sonographic patterns such as
echogenicity and textures are modeled and used to recognize incoming cases [27]. A previous
study demonstrated that the likelihood estimation of a computer-aided diagnosis system can
be used to reduce observer variability [28]. In this research, consistently high performance in
the differentiation of breast tumors was achieved with the assistance of a computer-aided diag-
nosis system. For residents, the specificity of the breast tumor diagnosis was improved from
20% to 40% (p-value < 0.01) and the κ value from 0.09 to 0.53 (p-value < 0.001). For dedicated
breast imagers, the specificity was increased from 34% to 43% (p-value = 0.16) and the κ value
from 0.21 to 0.61 (p-value < 0.001).
In this study, a computer-aided tear classification (CTC) system based on the quantitative
intensity and texture features was proposed to classify rotator cuff tears in shoulder ultra-
sounds. The establishment of the CTC system is expected to provide consistent and objective
recommendations to junior physicians for clinical examinations.
surgeon who is also specialized in musculoskeletal ultrasound. The acquisition frequency was
8 MHz and depth of scanning was 4 cm with the focus of supraspinatus layer. All patients
selected to this study underwent only drug control or other conservative treatment before
ultrasound examination. The patients with post-operative intervention and recent injection
including hyaluronic acid, steroid and platelet-rich plasma (PRP) injection were excluded. The
patient population (n = 136) included 61 males and 75 females between 25 and 86 years of age,
with a mean age of 58.7 years, and 32 of them underwent bilateral shoulder ultrasound evalua-
tion due to bilateral shoulder symptoms. In 23 shoulders, because the shoulder morphology
appeared non-uniform, two ultrasound images of different long-axis cut were captured from a
shoulder. The image database was composed of 191 shoulder images including 89 images of
supraspinatus tendinopathy and 102 images of supraspinatus tear. Forty-two of 102 supraspi-
natus tears were full-thickness tears. Image selection and diagnosis of 191 shoulder ultrasound
images were all confirmed by an orthopedic shoulder surgeon and a physical medicine and
rehabilitation (PM&R) physician who was specialized in shoulder musculoskeletal ultrasound.
During the ultrasound examination, patients were placed in a standard sitting position with
shoulder extended, internally rotated, and a routine ultrasound procedures were followed. The
settings of the ultrasound scanner, such as time gain compensation, were consistent for all
patients. The acquired shoulder ultrasound images were stored as 8-bit images with gray-scale
values ranging from 0 to 255. According to previous meta-analysis study, diagnostic accuracy
of supraspinatus tears is high while performed by musculoskeletal radiologists and shoulder
orthopedic surgeons [20]. As the gold standard in the evaluation of the proposed CTC system,
an orthopedic shoulder surgeon and a PM&R physician who was specialize in shoulder muscu-
loskeletal ultrasound classified the lesions into 89 cases of supraspinatus tendon tendinopathy
and 102 cases of supraspinatus tear. Lesion areas were delineated by the same orthopedic
shoulder surgeon to enclose the necessary tissues while avoiding normal tendons. Image J was
the software used in showing ultrasound images and delineating. The delineation of supraspi-
natus lesion areas was also confirmed by the PM&R physician to obtain the consensus. Fig 1
provides ultrasound images of a case of tendinopathy and a case of a tear.
Feature extraction
The normal supraspinatus tendon is a convex beak-shaped hyperechoic structure in long-axis
view [29]. After the delineation of the lesion area in the supraspinatus tendon, the sonographic
appearance of the enclosed tissues was analyzed according to their echogenic properties.
Supraspinatus tendinopathies were irregular, and loss of homogeneous texture was observed.
Supraspinatus tears appeared to have irregular margins with hypoechoic areas [30, 31]. These
tears, particularly those associated with tendon thickness, can be full-thickness tears or partial-
thickness tears from the bursal to the articular surface [32]. Consequently, intensity and tex-
ture features were proposed to analyze the tissues enclosed in the delineated lesion area.
Intensity features. The gray-scale distribution of tissues in the lesion can be presented by
a probability distribution and form a histogram. The statistical characteristics of the histogram
can be quantified by the histogram moments [33, 34]. The quantitative moments provide
objective measures of the histogram, expressing the intensity difference between tendinopa-
thies and tears. These include the mean, variance, skewness, and kurtosis, namely, the first-,
second-, third-, and fourth-order central moments of a histogram. The mean, at the center of a
distribution, can be obtained by summarizing total pixel values and dividing the sum by the
pixel number. Variance indicates how uniform the gray-scale values are spread out. Skewness
estimates the symmetry of the value distribution such as a bias to one side or not. Taking nor-
mal distribution as a reference, kurtosis is a single-peaked shape with heavily weighted tails.
Fig 1. Supraspinatus tendon shown in ultrasound images. (a) A case of tendon tendinopathy. (b) A case of
supraspinatus tear. (c) and (d): The lesion contours of (a) and (b), respectively, which were delineated by a shoulder
orthopedic surgeon using ImageJ.
https://doi.org/10.1371/journal.pone.0212741.g001
occurrence matrices P = [p(i,j|d,θ)] were generated by scanning the pixels and their neighbors
in G. The matrix element P = [p(i,j|d,θ)] represented the frequency of two adjacent pixels with
values of i and j at a distance (d) and a direction (θ). Based on the matrix, 14 GLCM texture
features were calculated [36]. Fig 2 illustrates the distance d = 1 and the direction θ = 0˚, 45˚,
90˚, or 135˚ in the consideration of texture composition. d = 1 was used to better describe the
details of some lesions having less than 0.5 cm. For four co-occurrence matrices with different
angles which included all the combination of two adjacent neighbors, the means of the above
statistic features were calculated and extracted from the lesion areas and were combined with
intensity features in the classifier to express tissue characteristics, such as brightness, contrast,
and heterogeneity.
Statistical analysis
According to the sonographic appearance of supraspinatus tendinopathy and tear, the corre-
sponding quantitative features were proposed for classification. In the evaluation, several test
methods were used to determine whether these features can distinguish between tendinopa-
thies and tissue tears. First, a Kolmogorov-Smirnov test [37] was used to determine if the value
distribution of a feature was normal or not. Normally distributed features were then tested by
Student’s t-test [37], and non-normally distributed features were evaluated by a Mann-Whit-
ney U-test [37]. The resulting p-values <0.05 indicated whether a feature was statistically sig-
nificant in distinguishing between the supraspinatus tendinopathy and tear. To generate a
prediction model based on the combination of various quantitative features, different feature
combinations were evaluated in the logistic regression classifier by stepwise backward elimina-
tion to discover the most relevant combination of features with the lowest error rate. The equa-
tion of feature combination in the logistic regression classifier is:
Tear probability ¼ 1=ð1 þ expð 1 � ðf1 � C1 þ � � � fn � Cn constantÞÞÞ ð1Þ
where f1, fn are different features which multiply different C1, Cn as coefficients.
Due to the limited number of collected cases, the generalization ability of the prediction
model was assessed by leave-one-out cross-validation. A case picked from N cases (the total
number of collected cases) was used to test the model trained by the remaining N-1 cases.
Summarizing the classification result of the N cases accomplished the prediction performance.
In the performance evaluation, each case was given a probability indicating the likelihood
of tears. Cases with probability values � 0.5 were classified as tears, and those < 0.5 were clas-
sified as tendinopathy. According to the gold standard established by an orthopedic shoulder
Fig 2. The illustration of texture analysis considering neighboring pixel pairs of four directions: 0˚, 45˚, 90˚, and
135˚ and distance = 1.
https://doi.org/10.1371/journal.pone.0212741.g002
surgeon and a PM&R physician, the following five performance indices were generated: accu-
racy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value
(NPV). The tradeoffs between the sensitivity and specificity were calculated and illustrated
using a receiver operating characteristic (ROC) curve. Az, the area under the ROC curve, was
analyzed using ROCKIT software (C. Metz, University of Chicago, Chicago, IL, USA). Other
statistical testing methods were performed using SPSS (version 16 for Windows; SPSS, Chi-
cago, IL, USA).
Results
Tables 1 and 2 show whether the proposed image features, including intensity and texture fea-
tures, can be significant in tear classification. As a result, four intensity and 11 texture features
obtained a statistically significant p-value less than 0.001. After feature selection, the relevant
image features were selected and combined in the classifier to generate a prediction model.
Three performance results of the CTC system based on different feature sets are shown in
Table 3. After backward elimination, three of four intensity features including Mean, Skewness,
Kurtosis were selected and combined in the classifier. The intensity feature set attained an
accuracy of 91%, a sensitivity of 92%, and a specificity of 91%. For texture features, Correlation,
Information measure of correlation, and Inverse difference normalized were selected to be the
most relevant according to their combination performance. The texture feature set attained an
accuracy of 89%, a sensitivity of 89% and, a specificity of 89%. Benefiting from complementary
advantages, the combined intensity and texture feature sets including selected Mean, Kurtosis,
Inverse difference normalized, and Inverse difference moment achieved an accuracy of 92%,
which is better than using intensity and texture feature sets individually.
Fig 3 demonstrates a supraspinatus tear case that was misclassified by the texture feature set
but correctly classified by the combination of texture and intensity feature sets. In Fig 4, the
trade-offs between sensitivity and specificity are illustrated using ROC curves, with corre-
sponding Az values.
Discussion
The proposed CTC system based on intensity and texture features was established to interpret
tissue echogenicities of shoulder ultrasound images. The prediction model built by a logistic
regression classifier achieved an accuracy of 92% for identifying rotator cuff tears and tendino-
pathies. The high accuracy suggests that the proposed CTC system is useful for assessing the
presence of rotator cuff tears. The classification result was obtained via leave-one-out cross-
validation due to the limited cases. The accuracy presented in this study provides us a direction
that the proposed CTC system works well in tear classification while the morphology features
are useless for differentiation in the observation. With respect to the selected features, Tears
Table 1. The test results of intensity features using the Mann-Whitney U-test.
�
p-value<0.05 indicates a statistically significant difference.
https://doi.org/10.1371/journal.pone.0212741.t001
Table 2. The test results of texture features using student’s t-test (mean) or the Mann-Whitney U-test (median).
�
p-value<0.05 indicates a statistically significant difference.
https://doi.org/10.1371/journal.pone.0212741.t002
tend to be darker due to its higher value of mean intensity and centralized with higher kurtosis
value. Besides, high Inverse difference normalized, and Inverse difference moment mean the
gray-scale distribution is uniform and lacking variance.
The proposed CTC system focused on the diagnosis of supraspinatus tears. It is believed
that ultrasound has a high diagnostic accuracy for calcific tendinitis, although few studies have
assessed this [16, 38]. Calcific tendinitis has several forms, including microcalcification, large
soft calcification without acoustic shadow, and large hard calcification appearing as hypere-
choic convex with acoustic shadows. All forms of calcific tendinitis of supraspinatus are
detected with little difficulty through ultrasound examination [32]. A limitation of the clinical
practice of shoulder ultrasound is the inter-observer variability in diagnosing rotator cuff
tears. The inter-observer agreement of diagnosing rotator cuff tears, especially partial thickness
tears, is only poor to moderate [10] and should be improved for clinical application of the
shoulder ultrasound. Therefore, this study proposed a customized CTC system for the diagno-
sis of rotator cuff tears. Although how much improvement of inter-operator variability was
not presented in this study. Future experiment will be performed with scheduled people and
Table 3. The performance comparisons of intensity features, texture features, and the combination of both feature sets.
https://doi.org/10.1371/journal.pone.0212741.t003
Fig 3. Tear classification results with probabilities higher than 50% were classified to be tear. (a) a moderate
supraspinatus near full thickness tear with unobvious characteristics in the ultrasound image (hypoechoic area near the
tendon insertion indicated by a white arrow) was misclassified by the texture feature set (42%) but correctly classified
by the combination of texture and intensity feature sets (100%). (b) a small supraspinatus partial thickness tear at
bursal surface was misclassified by the intensity feature set (9%) but correctly classified by the combination of texture
and intensity feature sets (68%).
https://doi.org/10.1371/journal.pone.0212741.g003
time. A previous study used a portion of the lesion (30×60 pixels) for tissue characterization to
classify 80 rotator cuff lesions into groups and achieved 92.5% accuracy [39]. The numerous
features used in the experiment included the fractal dimension, the texture spectrum, the sta-
tistical feature matrix, the texture feature coding method, and the gray-level co-occurrence
matrix. Compared to the previous study, our system collected the whole lesion area of 102
tears rather than a sub-region of 20 tears to provide more representative distribution for the
evaluation. Using only intensity and texture features in this study is expected to be more effi-
ciency for clinical use. With the proposed system, promising recommendations can be pre-
sented to different operator professionals with varying experiences in identifying rotator cuff
tears and tendinopathy. The diagnosis of rotator cuff full-thickness tears influences the deci-
sion to undertake the surgical repair or arthroplasty.
In the literature, the assessment of full thickness rotator cuff tears is shown to have better
sensitivity and specificity compared to partial-thickness rotator cuff tears [20, 40, 41]. A contro-
versial and uncertain issue is whether the accuracy of ultrasound examinations for the assess-
ment of partial-thickness rotator cuff tears is sufficient [16, 42, 43]. With ultrasound
examinations, partial-thickness tears are diagnosed when there is a focal hypoechoic or
anechoic defect inside of the tendon, involving either the articular or the bursal surface and
Fig 4. The trade-offs between the sensitivity and specificity of the computer-aided tear classification system using
different feature sets are illustrated by receiver operating characteristic curves. (“Moment” is referred to intensity
model; “GLCM” is referred to texture model; and “Moment + GLCM” is referred to a combined model.)
https://doi.org/10.1371/journal.pone.0212741.g004
segmentation to verify the clinical usefulness of the proposed CTC system. Another limitation
is that the collected ultrasound images were generated using a consistent setting. The classifica-
tion result based on the intensity features extracted from these images achieved an accuracy of
91%. According to the result, using intensity features under consistent settings would be useful
in tear classification while the system is customized for a specified ultrasound scanner or setting.
In other situations, if the system targets at multi-center hospitals with various scanners, a cali-
bration procedure or more intensity-invariant features should be adopted [24].
MR arthrography is considered as the most sensitive and specific technique for diagnosis
according to the meta-analysis study when compared to ultrasound and MRI, and MRI and
ultrasound are comparable in accuracy [45]. Nevertheless, the indication and convenience of
MR arthrography is more limited than ultrasound and is an invasive procedure. Furthermore,
not all supraspinatus tears such as partial thickness tears should be treated with surgical or
arthroscopic surgery. Sampling error exists if surgical or arthroscopic findings were used as
the gold standard. According to previous meta-analysis study, diagnostic accuracy of supraspi-
natus tears is high while performed by musculoskeletal radiologists and shoulder orthopedic
surgeons [20], and the accuracy would be higher with the consensus of the experienced opera-
tors. For the reasons specified above, the gold standard was established by the consensus of the
experienced shoulder orthopedic surgeon and PM&R physician who specialize musculoskele-
tal ultrasound in this study.
This study proposed a CTC system which achieved a high accuracy (92%) in identifying
rotator cuff tears, including partial and full thickness tears, by analyzing tissue enclosed in the
lesion area. The proposed CTC system performed similar performance to the experienced
operators in terms of accuracy. According to a meta-analysis of diagnostic accuracy of ultra-
sound for rotator cuff tears [20], diagnostic accuracy may be greatest when operated by mus-
culoskeletal radiologists, followed by orthopedic surgeons. The pooled sensitivity and
specificity under the direction of musculoskeletal radiologists are both 95%. However, sensitiv-
ity is lowered to 46% when the diagnosis of partial thickness tears is given by general radiolo-
gists or radiographers [46, 47]. Consequently, the proposed CTC system can provide clinical
assistance for general radiologists or ultrasonographers who may not have comparable rates of
diagnostic accuracy as musculoskeletal radiologists or orthopedic shoulder surgeons [20].
Additional experiments are needed to explore the clinical application of the proposed CTC
system. In particular, future research should examine how to utilize the CTC system in clinical
examinations to improve the performance of the observer. More specific, meaningful sono-
graphic findings may be needed to convince observers.
In conclusion, this CTC system based on intensity and texture features extracted from the
lesion area in shoulder ultrasound images achieved comparable accuracy in identifying rotator
cuff tears to musculoskeletal radiologists and orthopedic shoulder surgeons. The diagnostic sug-
gestions generated by the proposed CTC would be practical and promising in clinical assessments.
Supporting information
S1 Data. The delineated regions of rotator cuff tears.
(ZIP)
Acknowledgments
The authors would like to thank the Ministry of Science and Technology (MOST 107-2221-E-
004-013), New Taipei City Hospital (NTCH104-001) of Taiwan, the Republic of China, for
financially supporting this research.
Author Contributions
Conceptualization: Ruey-Feng Chang.
Data curation: Chung-Chien Lee.
Formal analysis: Chung-Chien Lee.
Funding acquisition: Chung-Ming Lo.
Investigation: Chung-Ming Lo.
Methodology: Chung-Ming Lo.
Resources: Chung-Chien Lee.
Validation: Ruey-Feng Chang, Chung-Ming Lo.
Writing – original draft: Chung-Ming Lo.
Writing – review & editing: Ruey-Feng Chang, Chung-Chien Lee.
References
1. Mitchell C, Adebajo A, Hay E, Carr A. Shoulder pain: diagnosis and management in primary care. Bmj.
2005; 331(7525):1124–8. https://doi.org/10.1136/bmj.331.7525.1124 PMID: 16282408
2. Luime JJ, Koes BW, Hendriksen IJ, Burdorf A, Verhagen AP, Miedema HS, et al. Prevalence and inci-
dence of shoulder pain in the general population; a systematic review. Scand J Rheumatol. 2004; 33
(2):73–81. PMID: 15163107.
3. Hidalgo-Lozano A, Fernández-de-las-Peñas C, Alonso-Blanco C, Ge H-Y, Arendt-Nielsen L, Arroyo-
Morales M. Muscle trigger points and pressure pain hyperalgesia in the shoulder muscles in patients
with unilateral shoulder impingement: a blinded, controlled study. Experimental brain research. 2010;
202(4):915–25. https://doi.org/10.1007/s00221-010-2196-4 PMID: 20186400
4. Bigliani LU, Levine WN. Current concepts review-subacromial impingement syndrome. J Bone Joint
Surg Am. 1997; 79(12):1854–68. PMID: 9409800
5. Lewis J, Tennent T, MacAuley D, Best T. How effective are diagnostic tests for the assessment of rota-
tor cuff disease of the shoulder. Evidence Based Sports Medicine. 2007:327–60.
6. Yamamoto A, Takagishi K, Osawa T, Yanagawa T, Nakajima D, Shitara H, et al. Prevalence and risk
factors of a rotator cuff tear in the general population. Journal of Shoulder and Elbow Surgery. 2010; 19
(1):116–20. https://doi.org/10.1016/j.jse.2009.04.006 PMID: 19540777
7. Milosavljevic J, Elvin A, Rahme H. Ultrasonography of the rotator cuff: a comparison with arthroscopy in
one-hundred-and-ninety consecutive cases. Acta Radiologica. 2005; 46(8):858–65. PMID: 16392611
8. Naqvi GA, Jadaan M, Harrington P. Accuracy of ultrasonography and magnetic resonance imaging for
detection of full thickness rotator cuff tears. International journal of shoulder surgery. 2009; 3(4):94.
https://doi.org/10.4103/0973-6042.63218 PMID: 20532011
9. Murphy RJ, Daines MT, Carr AJ, Rees JL. An independent learning method for orthopaedic surgeons
performing shoulder ultrasound to identify full-thickness tears of the rotator cuff. JBJS. 2013; 95
(3):266–72.
10. O’Connor PJ, Rankine J, Gibbon WW, Richardson A, Winter F, Miller JH. Interobserver variation in
sonography of the painful shoulder. J Clin Ultrasound. 2005; 33(2):53–6. https://doi.org/10.1002/jcu.
20088 PMID: 15674840.
11. Park HB, Yokota A, Gill HS, El Rassi G, McFarland EG. Diagnostic accuracy of clinical tests for the dif-
ferent degrees of subacromial impingement syndrome. J Bone Joint Surg Am. 2005; 87(7):1446–55.
https://doi.org/10.2106/JBJS.D.02335 PMID: 15995110.
12. Smith T, Back T, Toms A, Hing C. Diagnostic accuracy of ultrasound for rotator cuff tears in adults: a
systematic review and meta-analysis. Clinical radiology. 2011; 66(11):1036–48. https://doi.org/10.
1016/j.crad.2011.05.007 PMID: 21737069
13. van Holsbeeck MT, Kolowich PA, Eyler WR, Craig JG, Shirazi KK, Habra GK, et al. US depiction of par-
tial-thickness tear of the rotator cuff. Radiology. 1995; 197(2):443–6. https://doi.org/10.1148/radiology.
197.2.7480690 PMID: 7480690
14. Alasaarela E, Leppilahti J, Hakala M. Ultrasound and operative evaluation of arthritic shoulder joints.
Annals of the rheumatic diseases. 1998; 57(6):357–60. PMID: 9771210
15. Read JW, Perko M. Shoulder ultrasound: diagnostic accuracy for impingement syndrome, rotator cuff
tear, and biceps tendon pathology. J Shoulder Elbow Surg. 1998; 7(3):264–71. PMID: 9658352.
16. Martin-Hervas C, Romero J, Navas-Acien A, Reboiras JJ, Munuera L. Ultrasonographic and magnetic
resonance images of rotator cuff lesions compared with arthroscopy or open surgery findings. J Shoul-
der Elbow Surg. 2001; 10(5):410–5. https://doi.org/10.1067/mse.2001.116515 PMID: 11641696.
17. Roberts CS, Walker JA 2nd, Seligson D. Diagnostic capabilities of shoulder ultrasonography in the
detection of complete and partial rotator cuff tears. Am J Orthop (Belle Mead NJ). 2001; 30(2):159–62.
PMID: 11234944.
18. Teefey SA, Hasan SA, Middleton WD, Patel M, Wright RW, Yamaguchi K. Ultrasonography of the rota-
tor cuff. A comparison of ultrasonographic and arthroscopic findings in one hundred consecutive cases.
J Bone Joint Surg Am. 2000; 82(4):498–504. PMID: 10761940.
19. Miller D, Frost A, Hall A, Barton C, Bhoora I, Kathuria V. A ’one-stop clinic’ for the diagnosis and man-
agement of rotator cuff pathology: Getting the right diagnosis first time. Int J Clin Pract. 2008; 62
(5):750–3. https://doi.org/10.1111/j.1742-1241.2007.01682.x PMID: 18412933.
20. Smith TO, Back T, Toms AP, Hing CB. Diagnostic accuracy of ultrasound for rotator cuff tears in adults:
a systematic review and meta-analysis. Clin Radiol. 2011; 66(11):1036–48. https://doi.org/10.1016/j.
crad.2011.05.007 PMID: 21737069.
21. Le Corroller T, Cohen M, Aswad R, Pauly V, Champsaur P. Sonography of the painful shoulder: role of
the operator’s experience. Skeletal radiology. 2008; 37(11):979–86. https://doi.org/10.1007/s00256-
008-0539-z PMID: 18651142
22. Middleton WD, Teefey SA, Yamaguchi K. Sonography of the rotator cuff: analysis of interobserver vari-
ability. American Journal of Roentgenology. 2004; 183(5):1465–8. https://doi.org/10.2214/ajr.183.5.
1831465 PMID: 15505321
23. Lo C-M, Lai Y-C, Chou Y-H, Chang R-F. Quantitative breast lesion classification based on multichannel
distributions in shear-wave imaging. Computer methods and programs in biomedicine. 2015; 122
(3):354–61. https://doi.org/10.1016/j.cmpb.2015.09.004 PMID: 26421696
24. Lo C-M, Moon WK, Huang C-S, Chen J-H, Yang M-C, Chang R-F. Intensity-invariant texture analysis
for classification of bi-rads category 3 breast masses. Ultrasound in medicine & biology. 2015; 41
(7):2039–48.
25. Lo C-M, Chen Y-P, Chang Y-C, Lo C, Huang C-S, Chang R-F. Computer-aided strain evaluation for
acoustic radiation force impulse imaging of breast masses. Ultrasonic imaging. 2014; 36(3):151–66.
https://doi.org/10.1177/0161734613520599 PMID: 24894867
26. Flores WG, de Albuquerque Pereira WC, Infantosi AFC. Improving classification performance of breast
lesions on ultrasonography. Pattern Recognition. 2015; 48(4):1125–36.
27. Matta TTd, Pereira WCdA, Radaelli R, Pinto RS, Oliveira LFd. Texture analysis of ultrasound images is
a sensitive method to follow-up muscle damage induced by eccentric exercise. Clinical Physiology and
Functional Imaging. 2017.
28. Singh S, Maxwell J, Baker JA, Nicholas JL, Lo JY. Computer-aided classification of breast masses: per-
formance and interobserver variability of expert radiologists versus residents 1. Radiology. 2011; 258
(1):73–80. https://doi.org/10.1148/radiol.10081308 PMID: 20971779
29. Petranova T, Vlad V, Porta F, Radunovic G, Micu MC, Nestorova R, et al. Ultrasound of the shoulder.
Med Ultrason. 2012; 14(2):133–40. PMID: 22675714
30. Allen GM, Wilson DJ. Ultrasound of the shoulder. Eur J Ultrasound. 2001; 14(1):3–9. PMID: 11567849.
31. Vlychou M, Dailiana Z, Fotiadou A, Papanagiotou M, Fezoulidis IV, Malizos K. Symptomatic partial rota-
tor cuff tears: diagnostic performance of ultrasound and magnetic resonance imaging with surgical cor-
relation. Acta Radiol. 2009; 50(1):101–5. https://doi.org/10.1080/02841850802600764 PMID:
19052931.
32. Beggs I, editor Shoulder ultrasound. Seminars in Ultrasound, CT and MRI; 2011: Elsevier.
33. Groeneveld RA, Meeden G. Measuring skewness and kurtosis. The Statistician. 1984:391–9.
34. Baek HJ, Kim HS, Kim N, Choi YJ, Kim YJ. Percent change of perfusion skewness and kurtosis: a
potential imaging biomarker for early treatment response in patients with newly diagnosed glioblasto-
mas. Radiology. 2012; 264(3):834–43. https://doi.org/10.1148/radiol.12112120 PMID: 22771885
35. Haralick RM, Shanmugam K. Textural features for image classification. IEEE Transactions on systems,
man, and cybernetics. 1973;(6):610–21.
36. Mou WY, Guo DM, Liu H, Zhang P, Shao Y, Wang SW, et al. Staging liver fibrosis by analysis of non-lin-
ear normalization texture in gadolinium-enhanced magnetic resonance imaging. Biomedical Physics &
Engineering Express. 2015; 1(4):045012.
37. Field AP. Discovering statistics using SPSS, 3rd ed. Los Angeles: SAGE Publications; 2009.
38. Kayser R, Hampf S, Pankow M, Seeber E, Heyde CE. [Validity of ultrasound examinations of disorders
of the shoulder joint]. Ultraschall Med. 2005; 26(4):291–8. https://doi.org/10.1055/s-2005-858525
PMID: 16123923.
39. Horng M-H, Chen S-M. Multi-class classification of ultrasonic supraspinatus images based on radial
basis function neural network. J Med Biol Eng. 2009; 29(5):242–50.
40. Middleton WD, Teefey SA, Yamaguchi K. Sonography of the rotator cuff: analysis of interobserver vari-
ability. AJR Am J Roentgenol. 2004; 183(5):1465–8. https://doi.org/10.2214/ajr.183.5.1831465 PMID:
15505321.
41. Teefey SA, Rubin DA, Middleton WD, Hildebolt CF, Leibold RA, Yamaguchi K. Detection and quantifica-
tion of rotator cuff tears. Comparison of ultrasonographic, magnetic resonance imaging, and arthro-
scopic findings in seventy-one consecutive cases. J Bone Joint Surg Am. 2004; 86-A(4):708–16. PMID:
15069134.
42. Mitchell C, Adebajo A, Hay E, Carr A. Shoulder pain: diagnosis and management in primary care. Bmj.
2005; 331(7525):1124–8. https://doi.org/10.1136/bmj.331.7525.1124 PMID: 16282408; PubMed Cen-
tral PMCID: PMC1283277.
43. Moosmayer S, Heir S, Smith HJ. Sonography of the rotator cuff in painful shoulders performed without
knowledge of clinical information: results from 58 sonographic examinations with surgical correlation. J
Clin Ultrasound. 2007; 35(1):20–6. https://doi.org/10.1002/jcu.20286 PMID: 17149764.
44. Rutten MJ, Jager GJ, Blickman JG. From the RSNA refresher courses: US of the rotator cuff: pitfalls,
limitations, and artifacts. Radiographics. 2006; 26(2):589–604. https://doi.org/10.1148/rg.262045719
PMID: 16549619.
45. de Jesus JO, Parker L, Frangos AJ, Nazarian LN. Accuracy of MRI, MR arthrography, and ultrasound in
the diagnosis of rotator cuff tears: a meta-analysis. American Journal of Roentgenology. 2009; 192
(6):1701–7. https://doi.org/10.2214/AJR.08.1241 PMID: 19457838
46. Brenneke SL, Morgan CJ. Evaluation of ultrasonography as a diagnostic technique in the assessment
of rotator cuff tendon tears. The American journal of sports medicine. 1992; 20(3):287–9. https://doi.
org/10.1177/036354659202000309 PMID: 1636859
47. Takagishi K, Makino K, Takahira N, Ikeda T, Tsuruno K, Itoman M. Ultrasonography for diagnosis of
rotator cuff tear. Skeletal radiology. 1996; 25(3):221–4. PMID: 8741055