Secondary Logo

Journal Logo

SECTION II: ORIGINAL ARTICLES: Shoulder

Do Shoulder Vibration Signals Vary Among Asymptomatic Volunteers?

Kargus, Robert*; Bahu, Maher; Kahugu, Mark*; Martin, Sidney; Atkinson, Patrick*†

Author Information
Clinical Orthopaedics and Related Research: March 2007 - Volume 456 - Issue - p 103-109
doi: 10.1097/BLO.0b013e31802c3423

Abstract

Complaints of pain and/or dysfunction at the shoulder or elbow are common, accounting for one in eight problems treated by orthopaedic surgeons.36 Tests to diagnose the source of the pain and/or dysfunction include physical examinations,3,6,20,23,27 imaging studies,1,37,39 and invasive operative techniques.37,39,40 Although arthroscopy typically is considered a diagnostic standard for shoulders with intraarticular or subacromial lesions,37,40 sensitivities of approximately 90% have been achieved via imaging studies.1,37 In pair-matched studies comparing imaging studies and physical examinations, the examinations provided equivocal19 or reduced1 sensitivity. Although challenges in diagnosis via physical examination can be related to the specific lesion,12 the sensitivity of the examination can increase with the expertise of the clinician.19

One potential low-cost, noninvasive diagnostic area of study not applied to the shoulder is analyzing the vibrations emanating from the joint. The technique has been used to study vibrations emanating from articular joints including the knee, hip, and temporomandibular regions.9,10,13,14,16,17,21,28 These vibrations, collectively termed vibroarthrographic sounds or signals,42 are known to occur in the shoulder. They have been described as clunks,11 excessive crepitus,15 or snapping20 and are associated with painful joint function. Shoulder vibroarthrographic signals can be elicited from the joint during active or passive motion15,20 or during a physician-assisted test.11 It has been suggested these signals are associated with lesions or morphologic changes,15,38 thus suggesting a diagnostic benefit of analyzing its source.9,10,21

Although vibroarthrographic signals at the shoulder can be appreciated manually during clinical examination,12,15,20 additional levels of scrutiny are possible with electronic capture, storage, and subsequent analysis using statistical tests and tools such as neural networks.14,21,28 Although there are reports of the use of instrumented stethoscopes,2,41 more recent studies have shown the superior performance of skin-mounted accelerometers.14,21,28 Vibration data recruited using this method have been used to study asymptomatic joints in subjects,17 diagnose lesions (meniscal, ligamentous, arthritic),14,21,28 and predict the in vivo stability and wear of orthopaedic implants.8 Diagnostic sensitivities of 69% to 100% and specificities of 75% to 100% have been reported in the knee, hip, and jaw.13,14,17,21,24,28

A common vibroarthrographic method for diagnosing joint lesions is to recruit signals from a group of asymptomatic subjects followed by acquisition of additional signals from symptomatic joints grouped by disorder.24 Subsequent signals from a patient being diagnosed are compared with the known signals for classification. However, the findings may vary because limb dominance may influence the vibroarthrographic signals emanating from a joint. If signal variations were related to limb dominance, this likely would necessitate developing unique dominant and nondominant limb classifications. Although there are limited vibroarthrographic data for within-subject comparisons of a given joint, studies of the hip have had little variation9-approximately 3%.16 These subtle variations may be related to the tendency for varied articular features of the joint in the dominant limb as has been observed in the knee.4 There is also the tendency for nonoverhand-throwing subjects to have larger scapulae on the dominant side.38

Given the lack of clarity on the role of limb dominance, we hypothesized there would be no intrasubject differences for a given joint motion as a function of limb dominance.

MATERIALS AND METHODS

We studied vibroarthrographic signals from dominant and nondominant shoulders in 20 asymptomatic volunteers as the shoulder was mobilized through 21 different motions described below (12 active range of motion and 9 physician-assisted tests). We then compared the data between each subject's shoulders to determine if the signals were sufficiently different to be classified as recruited from a dominant or nondominant limb. We included subjects 18- to 24-years-old based on their relevant medical history and participation in certain sports. We avoided the recruitment of older subjects (>55 years) since they present with more substantial pathologies such as rotator cuff tears.29,30,32 There was a concern that such defects, whether symptomatic or not, could elicit errant vibroarthrographic signals. We also excluded subjects with a history of multidirectional instability, prior injury, chronic shoulder pain, or involvement in overhand throwing sports. We did not take radiographs so there was no attempt to exclude subjects based on the presence of an os acromiale. Our power analysis was based on surrogate data taken from the literature16 which represents the body of work studying within subject joint vibration characteristics in asymptomatic adults. That data was based on a discrepancy measurement of vibrations transmitted across both hips without hip motion and takes the modified ratio of right to left hip data. Kwong et al present an average ± one standard deviation for a dimensionless ratio (right versus left frequencies) with 8 replicate measures per subject.16 An example of their maximal variability is presented at 100 Hz in which they report a mean ratio of 4.54 ± 4.35 considering all individual replicate measures; we used all data points since all presented with different amounts of data dispersion. We established β > 0.8 as our standard to establish the minimum sample size. Using the variability we computed from the data of Kwong et al, the power analysis using a paired t-test indicated 10 subjects would yield a power of β = 0.835; 9 subjects a power of β = 0.78. The sample size of 20 subjects (40 shoulders) in the current study was chosen to allow an equal sample of genders, accommodate statistical recommendations of preferred sample sizes, and to limit the dispersion of data recruited from a single shoulder per subject.26,36 The study protocol was reviewed and approved by an Institutional Review Board, and all subjects completed the necessary informed consent and HIPAA forms. They were also compensated for their time.

Because the velocity at which a joint is mobilized influences joint signal intensity, a visual cueing method has been developed to aid subjects during active tests in attaining a given target velocity.9,10 Each volunteer mimicked a video projection of a person mobilizing their shoulder for a given motion and velocity (Fig 1). In a pilot test of 12 volunteers, a skin-mounted bi-axis electrogoniometer (Biometrics SG150, Lady Smith, VA) was affixed over the superior surface of the shoulder in the coronal plane. Two velocities of 10°/second and 60°/second were studied. The volunteers achieved the target velocity within a standard deviation of 9%.

Fig 1
Fig 1:
A skin mounted accelerometer was used to record vibroarthrographic signals emanating from the shoulder region during a series of physician-assisted and active ROM maneuvers. As shown, subjects were directed regarding speed and range of their active motions by mimicking a video projection of a given motion.

All shoulder vibroarthrographic signals were acquired with one skin-mounted, triaxial accelerometer (frequency range, 0.4-3000 Hz; sensitivity, 1 V/g) (Dytran 3233A, Chatsworth, CA) at a sampling frequency of 10 kHz. We used data from the channel oriented perpendicular to the skin surface. The sensor's frequency range was anticipated to contain the relevant physiologic joint vibrations based on previous vibroarthrographic studies of the hip, knee, and temporomandibular joints.9,10,21,28 The sensitivity was anticipated to have sufficient resolution because McCoy et al21 acquired signals from the knee in the 0.5 g range for low amplitude signals associated with crepitus. For placement of the sensor, Nokes et al25 established the importance of positioning the accelerometer over a bony prominence adjacent to the joint to increase the vibration signal transmitted to the accelerometer. Therefore, we attached the accelerometer to the skin overlying the acromion with one of the accelerometer's sensitive axes oriented normal to the skin.25 This site was chosen for its proximity to the glenohumeral joint and its ease of identification via bony palpation. We affixed the accelerometer via two methods to limit vibration artifact from sensor migration or cable motion.23 First, we attached the sensor using a biocompatible, high-tack, spray-on adhesive commonly used to secure prosthetic devices to the skin. We then molded a urethane compound to fit over the accelerometer and extend beyond the sensor's periphery (Fig 1). We attached medical tape over the construct for added security.

After attaching the accelerometer, we used the visual cueing method to direct the volunteers to actively rotate their shoulder in six directions: abduction, adduction, flexion, extension, internal rotation, and external rotation. Each motion was executed at two different velocities (fast at 60°/second and slow at 10°/second) for a total of 12 active motions. The fast rate was chosen as the most comfortable velocity for swinging knees quickly.9,10 Alternately, pilot testing showed the slowest speed readily attainable for long duration motions such as abduction was 10°/second. Abduction was initiated with the upper extremity at the subject's side and moved through 180°. Adduction was performed with the shoulder first flexed 90° in the scapular plane followed by 45° adduction. Flexion was executed by holding the arm at the side and flexing the shoulder 180° in the scapular plane. Extension was initiated with the arm at the side while the subject extended the shoulder rearward through a range of approximately 70°. Internal rotation was executed with the arm behind the back while the subject reached up his or her back; external rotation was executed by holding the arm at the subject's side with the elbow flexed at 90°, and then rotating the extremity through 75°. There was no attempt at scapular stabilization when measuring total glenohumeral plus scapulothoracic motion. Nine physician-assisted tests were selected to complete the shoulder physical examination (Table 1).6,27,33 The duration of each of the 21 tests varied from approximately 2 seconds for fast motions and 18 seconds for slow motions such as abduction.

TABLE 1
TABLE 1:
Physician-assisted Shoulder Examination Tests

The data from the 21 tests the signals were filtered before analysis. The filtering and analysis techniques were applied to the voltage output values of the accelerometer channel oriented perpendicular to the skin, not on the scaled signal (ie, in units of m/s2). We performed filtering with a Fast Fourier Transform algorithm which excluded frequencies less than 5 Hz, from 58 to 62 Hz, and greater than 1000 Hz using commercially available software (Matlab v. 7, The MathWorks Inc, Natick, MA). We excluded frequencies less than 5 Hz to eliminate gross motion of the shoulder and muscle contraction interference.42 The filter from 58 to 62 Hz eliminated 60 Hz AC power artifact common to lighting.23 We eliminated frequencies greater than 1000 Hz after initial reviews of shoulder vibroarthrographic data revealed negligible time-varying data in excess of 1000 Hz.

To prepare the time-varying acceleration data for statistical comparison, a wavelet data reduction was applied to the filtered data. Such use of wavelets is common with filtered vibroarthrographic signals as they typically contain time-varying data that appear as complicated sine waves with varying amplitudes.9,10,13,14,31 Wavelet analysis is attractive by objectively reducing the complicated, nonrepeating signals into a short series of numbers that describe the original data.7 Based on an analysis of the data from the current study and the direction from prior investigations,9,10,21 the following technical issues were addressed: a continuous wavelet transform was used to calculate time-varying coefficients for five acceleration frequencies (5, 30, 75, 300, 1000 Hz) using a Gaussian waveform of single replication. The five wavelet coefficient time histories corresponding to the five frequencies were segregated into five equally spaced temporal windows over the duration of the particular test. This normalized the amount of data for the different tests which required 2 to 18 seconds to complete. We then calculated the root-mean-squared (RMS) of coefficients in a given window, parameterizing the original vibroarthrographic signal into a set of 25 numerical values for each shoulder for each test.28,35 The final data analysis was limited to 17 subjects (eight males, nine females), as one or more vibroarthrographic signal files in three subjects were errantly overwritten, yielding an incomplete record.

We used Wilcoxon rank sum tests for each of the 25 numerical values describing a shoulder for a given test. Significance was set at p < 0.05. Each of the 25 numerical values for the 21 different physical examination tests were compared within subject for the 17 volunteers in 525 (25 multiplied by 21) statistical tests. The Wilcoxon rank sum test was selected over other parametric two-sample tests because 93% of the data from the different physical tests were not normally distributed.

Because of the complexity of the reduced data (common to vibroarthrographic studies), we used a neural network tool to identify similarities or differences in joint signals.5,28,35 This technique relies on regression statistics to establish a model by submitting data from a randomly selected subset of the subjects in which the data are known to have been acquired from a dominant or nondominant shoulder. This training of the network then allows data from the remaining unknown shoulders to be submitted for classification by the neural network as either from a dominant or nondominant shoulder. In the current study, neural network classification rates represent the sensitivity of vibroarthrography in identifying a signal as being from a dominant or nondominant shoulder. This method has the benefit of collectively considering all 25 values from each subject's shoulder versus the paired testing described above which is limited to comparisons of one value at a time. We used commercially available software (NeuroSolutions, NeuroDimension Inc, Gainesville, FL) to establish regression models of dominant and nondominant data using vibroarthrography signals from 12 randomly selected subjects (24 shoulders). Data from the remaining five subjects (10 shoulders) were submitted to the network without regard to limb dominance. The classification performance of the network then was assessed.

RESULTS

The within-subject assessments of shoulder vibration signals showed little difference based on limb dominance when considering all joint motions. First, this was observed via statistical comparisons which revealed 21 data points or 4% of all 525 data points were significant (Table 2). Differences ranged from 0.001 < p ≤ 0.049 and were observed in all windows and all pseudofrequencies. For the 21 statistically different comparisons, the average (RMS) coefficient values describing the shoulders from the dominant limbs were greater than the data from the nondominant side. Eleven of the 21 significant differences were from the following tests: slow ROM (two tests), fast ROM (four tests), and physician-assisted (five tests). The majority of the differences were seen in windows 2 to 5 in pseudofrequencies 1 to 4. Findings similar to the statistical comparisons also were seen by the neural network which correctly classified a shoulder vibroarthrographic signal as dominant or nondominant an average of 50% and 48% of the time, respectively (Table 3). Classification rates varied from 0% to 100% and 20% to 100% for the dominant and nondominant limbs, respectively. The best classification for active tests was achieved for the nondominant shoulder signals during internal rotation. For physician-assisted tests, dominant shoulder signals from Neer's impingement sign and Yergeson's test yielded 100% correct classification. The poorest classification was observed for fast and slow active flexion in which all dominant shoulders were misclassified as nondominant.

TABLE 2
TABLE 2:
Statistical Comparison of the 525 Data Points
TABLE 3
TABLE 3:
Neural Network Classification of Shoulder Dominance

DISCUSSION

Our objective was to study vibroarthrographic signals emanating from dominant and nondominant shoulders during active ROM and physician-assisted physical examination tests. Because of the small morphologic differences in articular joints based on limb dominance, we hypothesized there would be no difference in vibroarthrographic signals between the shoulders.

We note several limitations inherent to this study, the first of which being the study population. The presence or absence of joint lesions in asymptomatic patients is known to be related to the age of the subject. For example, the prevalence of partial or full-thickness asymptomatic rotator cuff tears increases substantially with age.30,32 As such, the increased compliance of such a joint complex may yield vibroarthrographic signals which are appreciably different than in our study. Asymptomatic subjects aged 19 to 39, 40 to 60, and older than 60 years presented with asymptomatic rotator cuff lesions in 4%, 4%, and 54% of subjects, respectively.32 Thus, the data from the current study yield an appropriate baseline representing asymptomatic subjects in their third to sixth decades with regard to the rotator cuff. However, additional studies are needed for asymptomatic subjects known to present with a rotator cuff tear leading to an additional asymptomatic category which likely would be comprised of older subjects. Additional lesions unique to the shoulder which may be symptomatic or asymptomatic also bear additional scrutiny. This would include an analysis of patients presenting with an os acromiale. Although patients were not screened for the presence of os acromiale, we believed the presence of such a mobile feature in an otherwise asymptomatic subject would not yield excessive vibration. However, this must be confirmed in future studies. Attributing a vibration signal to a specific anatomic structure likely would expand the usefulness of vibroarthrography. This has been accomplished in the knee by attaching multiple sensors to bony prominences. The sensors receiving the strongest signals suggested particular vibrations were most likely emanating from the tissues in closest proximity to a given sensor. Application of this methodology to the shoulder could help isolate vibrations as either emanating from the scapulothoracic joint versus the glenohumeral region, for example.

An additional limitation in our study is the large number of the average (RMS) coefficient values that were not significant. This finding necessitates an appreciation of the power. A retrospective analysis of the power of the current study would rely on a parametric statistic which would be inappropriate because the majority of data in the current study are not normal.18 However, our sample size was based on a power analysis using the data from Kwong et al.16 Based on that analysis and the large amount of within-subject data that were recruited, we believe the majority of nonsignificant statistical comparisons are supported.26,36

Limited vibroarthrographic data acquired from dominant and nondominant limbs in adult, asymptomatic joints are available. One notable exception is the bilateral study of hips16 in which adult volunteers were studied as to the native symmetry of transarticular vibration transmission. Similar to our findings, they reported small intrasubject variations of approximately 3%, thus supporting how there is little variation in vibration transmission across the asymptomatic hip. Similar testing on young children and neonates yielded greater variability in the data; the exact cause for this requires additionalstudy.16

The utility in classifying an unknown signal with neural networks has been shown in various medical applications, namely articular joint sounds, EEG, and tumor classification.5,28,35 Radke et al28 reported 87% accuracy in correctly diagnosing the presence or absence of temporomandibular joint disorder using such a network. However, the apparent lack of differences observed from the paired t tests in our study was associated with poor classification performance. The neural network correctly classified an average of 50% of the shoulders. This is no different than expected by randomly selecting from a variable with two levels. The findings of the neural network may be more helpful in identifying similarities (or differences) because it can collectively compare a greater amount of data than traditional statistics. Because each vibroarthrographic signal was reduced from a complex waveform to 25 individual numbers, the within-subject statistical comparisons were limited to one value per comparison. This precluded an overall analysis of all data acquired from a subject for a given physical examination test leading to potentially erroneous conclusions based on individual data point by data point comparisons. Conversely, the neural network used all 25 values from a given test to determine its best classification as either coming from a dominant or nondominant shoulder.

Vibroarthrographic signals of dominant and nondominant shoulders in our group of healthy volunteers were significantly different in only a small number of comparisons. These small differences detected from the two-sample tests and the low sensitivity of the neural network suggests our data may be used as a reference for asymptomatic patients 19 to 60 years old. In six of 42 classifications, high classification rates (80-100% of the five dominant or nondominant shoulders) were noted for data associated with three each of ROM and physician-assisted tests. This high rate of classification suggests the network was largely successful in differentiating shoulder dominance for a limited group of shoulder motions. Alternately, low classification rates of 0% to 20% indicate a given dominant or nondominant shoulder is interpreted as the contralateral joint by the neural network. However, these observations may be overwhelmed by the presence of a lesion or degenerative changes. For example, additional studies investigating lesions which would be expected to produce amplified vibroarthrographic signals versus shoulders free of lesions would likely help in supporting our data.31 The variable performance of the neural network when attempting to distinguish between two similar joints (ie, within-subject comparisons of asymptomatic shoulders) might significantly improve when comparing a torn rotator cuff or loose arthroplasty component versus the asymptomatic data in our study. These latter joint problems would be expected to yield detectable differences via data capture and analysis as has been shown in worn or failed total knee arthroplasty components.8

Acknowledgments

We thank Dan Vancura, Dr. Jeff Hargrove, Dr. Tim Cameron, and Dr. Daniel Ludwigsen (Kettering University), and Dr. Dane Miller (Biomet Corporation) for support.

References

1. Ardic F, Kahraman Y, Kacar M, Kahraman MC, Findikoglu G, Yorgancioglu ZR. Shoulder impingement syndrome: relationships between clinical, functional, and radiologic findings. Am J Phys Med Rehabil. 2006;85:53-60.
2. Chu ML, Gradisar IA, Railey MR, Bowling GF. Detection of knee joint diseases using acoustic pattern recognition. J Biomech. 1976;9:111-114.
3. Dromerick AW, Kumar A, Volshteyn O, Edwards DF. Hemiplegic shoulder pain syndrome: interrater reliability of physical diagnosis signs. Arch Phys Med Rehabil. 2006;87:294-295.
4. Eckstein F, Muller S, Faber SC, Englmeier KH, Reiser M, Putz R. Side differences of knee joint cartilage volume, thickness, and surface area, and correlation with lower limb dominance: an MRI-based study. Osteoarthritis Cartilage. 2002;10:914-921.
5. Guler I, Ubeyli ED. Adaptive neuro-fuzzy inference system for classification of EEG signals using wavelet coefficients. J Neurosci Methods. 2005;148:113-121.
6. Hawkins RJ, Bokor DJ. Clinical evaluation of shoulder problems. In: Rockwood CA, Matsen FA 3rd, eds. The Shoulder. 2nd ed. Philadelphia, PA: WB Saunders Co; 1998:679-680.
7. Jaideva C. Goswami,Chan AK. Fundamentals of Wavelets. New York, NY: John Wiley & Sons; 1999:226-228.
8. Jiang CC, Lee JH, Yuan TT. Vibration arthrometry in the patients with failed total knee replacement. IEEE Trans Biomed Eng. 2000;47:219-227.
9. Kernohan WG, Beverland DE, McCoy GF, Hamilton A, Watson P, Mollan R. Vibration arthrometry. Acta Orthop Scand. 1990;61:70-79.
10. Kernohan WG, Cowie GH, Mollan RA. Vibration arthrometry in congenital dislocation of the hip. Clin Orthop Relat Res. 1991;272:167-174.
11. Kim SH, Park JC, Park JS, Oh I. Painful jerk test: a predictor of success in nonoperative treatment of posteroinferior instability of the shoulder. Am J Sports Med. 2004;32:1849-1855.
12. Kim SH, Park JS, Jeong WK, Shin SK. The Kim test: a novel test for posteroinferior labral lesion of the shoulder: a comparison to the jerk test. Am J Sports Med. 2005;33:1188-1192.
13. Krishnan S, Rangayyan RM. Automatic denoising of knee-joint vibration signals using adaptive time-frequency representations. Med Biol Eng Comput. 2000;38:2-8.
14. Krishnan S, Rangayyan RM, Bell GD, Frank CB. Adaptive time-frequency analysis of knee joint vibroarthrographic signals for non-invasive screening of articular cartilage pathology. IEEE Trans Biomed Eng. 2000;47:773-783.
15. Kuhn JE, Plancher KD, Hawkins RJ. Symptomatic scapulothoracic crepitus and bursitis. J Am Acad Orthop Surg. 1998;6:267-273.
16. Kwong KS, Huang X, Cheng JC, Evans JH. Acoustic transmission in normal human hips: structural testing of joint symmetry. Med Eng Phys. 2003;25:811-816.
17. Kwong KS, Huang X, Cheng JC, Evans JH. New technique for early screening of developmental dysplasia of the hip: pilot study. J Pediatr Orthop. 2003;23:347-351.
18. Lehmann EL. Nonparametrics: Statistical Methods Based on Ranks. New York, NY, McGraw Hill; 1975:76-81.
19. Litaker D, Pioro M, El Bilbeisi H, Brems J. Returning to the bedside: using the history and physical examination to identify rotator cuff tears. J Am Geriatr Soc. 2000;48:1633-1637.
20. Manske RC, Reiman MP, Stovak ML. Nonoperative and operative management of snapping scapula. Am J Sports Med. 2004;32:1554-1565.
21. McCoy GF, McCrea JD, Beverland DE, Kernohan WG, Mollan RA. Vibration arthrography as a diagnostic aid in diseases of the knee. J Bone Joint Surg Br. 1987;69:288-293.
22. Michener LA, Boardman ND, Pidcoe PE, Frith AM. Scapular muscle tests in subjects with shoulder pain and functional loss: reliability and construct validity. Phys Ther. 2005;85:1128-1138.
23. Mollan RA, Kernohan WG, Watters PH. Artefact encountered by the vibration detection system. J Biomech. 1983;16:193-199.
24. Moussavi ZM, Rangayyan RM, Bell GD, Frank CB, Ladly KO, Zhang YT. Screening of vibroarthrographic signals via adaptive segmentation and linear prediation modeling. IEEE Trans Biomed Eng. 1996;43:15-23.
25. Nokes L, Fairclought JA, Mintowt-Czyz WJ, Mackie I, Williams J. Vibration analysis of human tibia: the effect of soft tissue on the output from skin-mounted accelerometers. J Biomed Eng. 1984;6:223-226.
26. Nunally JC, Bernstein IH. Psychometric Theory. 3rd ed. New York, NY: McGraw Hill; 1994:244-245.
27. Park HB, Yokota A, Gill HS, El Rassi G, McFarland EG. Diagnostic accuracy of clinical tests for the different degrees of subacromial impingement syndrome. J Bone Joint Surg Am. 2005;87:1446-1455.
28. Radke JC, Ketcham R, Glassman B, Kull R. Artificial neural network learns to differentiate normal TMJs and nonreducing displaced disks after training on incisor-point chewing movements. Cranio. 2003;21:259-264.
29. Razmjou H, Holtby R, Myhr T. Gender differences in quality of life and extent of rotator cuff pathology. Arthroscopy. 2006;22:57-62.
30. Reilly P, Macleod I, Macfarlane R, Windley J, Emery RJ. Dead men and radiologists don't lie: a review of cadaveric and radiological studies of rotator cuff tear prevalence. Ann R Coll Surg Engl. 2006;88:116-121.
31. Shen Y, Rangayyan RM, Bell GD, Frank CB, Zhang YT, Ladly KO. Localization of knee joint cartilage pathology by multichannel vibroarthrography. Med Eng Phys. 1995;17:583-594.
32. Sher JS, Uribe JW, Posada A, Murphy BJ, Zlatkin MB. Abnormal findings on magnetic resonance images of asymptomatic shoulders. J Bone Joint Surg Am. 1995;77:10-15.
33. Snyder SJ, Karzel RP, Del Pizzo W, Ferkel RD, Friedman MJ. SLAP lesions of the shoulder. Arthroscopy. 1990;6:274-279.
34. Spiegler M. Knees have no warranties. Am Demographics. 1995;17:20-22.
    35. Stefanescu C, Boiculese LV, Rusu V, Costin M, Hountis D, Musat E. Cancer evolution assessment using artificial neural networks. Rev Med Chir Soc Med Nat Iasi. 2004;108:94-98.
    36. Tabachnik B, Fidell L. Using Multivariate Statistics. 3rd ed. New York, NY: HarperCollins Publishers Inc; 1996:132-133.
    37. Teefey SA, Rubin DA, Middleton WD, Hildebolt CF, Leibold RA, Yamaguchi K. Detection and quantification of rotator cuff tears: comparison of ultrasonographic, magnetic resonance imaging, and arthroscopic findings in seventy-one consecutive cases. J Bone Joint Surg Am. 2004;86:708-716.
    38. T'Jonck L. Lysens R, Grasse G. Measurements of scapular position and rotation: a reliability study. Physiother Res Int. 1996;1:148-158.
    39. Toyoda H, Ito Y, Tomo H, Nakao Y, Koike T, Takaoka K. Evaluation of rotator cuff tears with magnetic resonance arthrography. Clin Orthop Relat Res. 2005;439:109-115.
    40. Waldt S, Burkart A, Imhoff AB, Bruegel M, Rummeny EJ, Woertler K. Anterior shoulder instability: accuracy of MR arthrography in the classification of anteroinferior labroligamentous injuries. Radiology. 2005;237:578-583.
    41. Yao J, Zhang YT. Artefact encountered by the vibration detection system. J Biomech. 1983;16:193-199.
    42. Zhang YT, Rangayyan RM, Frank CB, Bell GD. Adaptive cancellation of muscle contraction interference in vibroarthrographic signals. IEEE Trans Biomed Eng. 1994;41:181-191.
    © 2007 Lippincott Williams & Wilkins, Inc.