• View in gallery
    Figure 1—

    Averaged spectrum (second derivative; dotted line) of a control group of synovial fluid samples and corresponding averaged difference spectrum (control minus affected [samples from horses with traumatic arthritis] spectra; solid line) in a study of use of infrared spectroscopy for diagnosis of traumatic arthritis in horses. Shaded areas (1,245 to 1,257 cm−1, 1,681 to 1,684 cm−1, and 1,691 to 1,694 cm−1) represent the optimal infrared regions for diagnostic classification of the spectra. Negative features in the second-derivative spectrum of the control group correspond to positive features (absorptions) in the original spectrum.

  • View in gallery
    Figure 2—

    Three-dimensional representation of spectral datasets preprocessed for classification and their division into calibration and validation sets in a study of use of infrared spectroscopy for diagnosis of traumatic arthritis in horses. Each observation is represented by the triplet of intensities in the regions 1,245 to 1,257 cm−1 (infrared region I), 1,681 to 1,684 cm−1 (infrared region II), and 1,691 to 1,694 cm−1 (infrared region III) for the second-derivative spectrum. Data are control spectra (open circles) and osteochodral fracture spectra (closed circles) in calibration set, control spectra (open triangles) and osteochondral fracture spectra (closed triangles) in validation set I, and normal control spectra in validation set II (diamonds).

  • 1

    Kidd JA, Fuller C, Barr ARS. Osteoarthritis in the horse. Equine Vet Educ 2001;13:160168.

  • 2

    Rossdale PD, Hopes R & Wingfield Digby NJ, et al. Epidemiological study of wastage among racehorses 1982 and 1983. Vet Rec 1985;116:6669.

  • 3

    McIlwraith CW. Diseases of joints, tendons, ligaments, and related structures. In: Stashak TS, ed. Adams' lameness in horses. 5th ed. Philadelphia: Lippincott Williams & Wilkins, 2002;459644.

    • Search Google Scholar
    • Export Citation
  • 4

    Park RD, Wrigley RH, Steyn PF. Equine diagnostic imaging. In: Stashak TS, ed. Adams' lameness in horses. 5th ed. Philadelphia: Lippincott Williams & Wilkins, 2002;185375.

    • Search Google Scholar
    • Export Citation
  • 5

    Van Pelt RW. Interpretation of synovial fluid findings in the horse. J Am Vet Med Assoc 1974;165:9195.

  • 6

    Kawcak CE, McIlwraith CW & Norrdin RW, et al. Clinical effects of exercise on subchondral bone of carpal and metacarpophalangeal joints in horses. Am J Vet Res 2000;61:12521258.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7

    Twardock AR. Equine bone scintigraphic uptake patterns related to age, breed, and occupation. Vet Clin North Am Equine Pract 2001;17:7594.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 8

    Kraft SL, Gavin P. Physical principle and technical considerations for equine computed tomography and magnetic resonance imaging. Vet Clin North Am Equine Pract 2001;17:115130.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9

    McIlwraith CW, Billinghurst RC, Frisbie DD. Current and future diagnostic means to better characterize osteoarthritis in the horse—routine synovial fluid analysis and synovial fluid and serum markers, in Proceedings. 47th American Association Equine Practice Annu Conv 2001;47:171179.

    • Search Google Scholar
    • Export Citation
  • 10

    Bertone AL, Palmer JL, Jones J. Synovial fluid cytokines and eicosanoids as markers of joint disease in horses. Vet Surg 2001;30:528538.

  • 11

    Trumble TN, Trotter GW & Oxford JR, et al. Synovial fluid gelatinase concentrations and matrix metalloproteinase and cytokine expression in naturally occurring joint disease in horses. Am J Vet Res 2001;62:14671477.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12

    Poole AR. Biochemical/immunochemical biomarkers of osteoarthritis: utility for prediction of incident or progressive osteoarthritis. Rheum Dis Clin North Am 2003;29:803818.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 13

    Shaw RA, Mantsch HH. Vibrational biospectroscopy: from plants to animals to humans. A historical perspective. J Mol Structure 1999;480/481:113.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 14

    Shaw RA, Mantsch HH. Infrared spectroscopy in clinical and diagnostic analysis. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;83102.

    • Search Google Scholar
    • Export Citation
  • 15

    Stuart B. Introduction. In: Infrared spectroscopy: fundamentals and applications. Chichester, England: John Wiley & Sons Ltd, 2004;113.

    • Search Google Scholar
    • Export Citation
  • 16

    Coates J. Interpretation of infrared spectra, a practical approach. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;1081510837.

    • Search Google Scholar
    • Export Citation
  • 17

    Dubois J, Shaw RA. IR spectroscopy in clinical and diagnostic applications. Anal Chem 2004;76:361A367A.

  • 18

    Jackson M, Mantsch HH. Infrared spectroscopy, ex vivo tissue analysis by. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;131156.

    • Search Google Scholar
    • Export Citation
  • 19

    Petrich W, Staib A & Otto M, et al. Correlation between the state of health of blood donors and the corresponding mid-infrared spectra of the serum. Vib Spectrosc 2002;28:117129.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 20

    Pizzi N, Choo LP & Mansfield J, et al. Neural network classification of infrared spectra of control and Alzheimer's affected tissue. Artif Intell Med 1995;7:6779.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21

    Jackson M, Mansfield JR & Dolenko B, et al. Classification of breast tumors by grade and steroid receptor status using pattern recognition analysis of infrared spectra. Cancer Detect Prev 1999;23:245253.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 22

    Canvin JM, Bernatsky S & Hitchon CA, et al. Infrared spectroscopy: shedding light on synovitis in patients with rheumatoid arthritis. Rheumatology (Oxford) 2003;42:7682.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 23

    Eysel HH, Jackson M & Nikulin A, et al. A novel diagnostic test for arthritis: multivariate analysis of infrared spectra of synovial fluid. Biospectroscopy 1997;3:161167.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24

    Shaw RA, Kotowich S & Eysel HH, et al. Arthritis diagnosis based upon the near-infrared spectrum of synovial fluid. Rheumatol Int 1995;15:159165.

  • 25

    Staib A, Dolenko B & Fink DJ, et al. Disease pattern recognition testing for rheumatoid arthritis using infrared spectra of human serum. Clin Chim Acta 2001;308:7989.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26

    Jackson M, Sowa MG, Mantsch HH. Infrared spectroscopy: a new frontier in medicine. Biophys Chem 1997;68:109125.

  • 27

    Shaw RA, Mantsch HH. Multianalyte serum assays from mid-IR spectra of dry films on glass slides. Appl Spectrosc 2000;54:885889.

  • 28

    Low-Ying S, Shaw RA & Leroux M, et al. Quantitation of glucose and urea in whole blood by mid-infrared spectroscopy of dry films. Vib Spectrosc 2002;28:111116.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 29

    Adams MJ. Feature selection and extraction. In: Chemometrics in analytical spectroscopy. Cambridge, England: The Royal Society of Chemistry, 2004;5595.

    • Search Google Scholar
    • Export Citation
  • 30

    Nikulin AE, Dolenko B & Bezabeh T, et al. Near-optimal region selection for feature space reduction: novel preprocessing methods for classifying MR spectra. NMR Biomed 1998;11:209216.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 31

    Khattree R, Naik DN. Discriminant analysis. In: Multivariate data reduction and discrimination with SAS software. Cary, NC: SAS Institute Inc, 2000;211345.

    • Search Google Scholar
    • Export Citation
  • 32

    Adams MJ. Pattern recognition II: supervised learning. In: Chemometrics in analytical spectroscopy. Cambridge, England: The Royal Society of Chemistry, 2004;129160.

    • Search Google Scholar
    • Export Citation
  • 33

    Zhou XH, Obuchowski NA, McClish DK. The design of diagnostic accuracy studies. In: Statistical methods in diagnostic medicine. New York: John Wiley & Sons, 2002;5799.

    • Search Google Scholar
    • Export Citation
  • 34

    Kawcak CE, McIlwraith CW & Norrdin RW, et al. The role of subchondral bone in joint disease: a review. Equine Vet J 2001;33:120126.

  • 35

    Stuart B. Spectral analysis. In: Infrared spectroscopy: fundamentals and applications. Chichester, England: John Wiley & Sons Ltd, 2004;4570.

    • Search Google Scholar
    • Export Citation
  • 36

    Clegg PD, Carter SD. Matrix metalloproteinase-2 and -9 are activated in joint diseases. Equine Vet J 1999;31:324330.

  • 37

    Dimock AN, Siciliano PD, McIlwraith CW. Evidence supporting an increased presence of reactive oxygen species in the affected equine joint. Equine Vet J 2000;32:439443.

    • Search Google Scholar
    • Export Citation
  • 38

    Frisbie DD, Ray CS & Ionescu M, et al. Measurement of synovial fluid and serum concentrations of the 846 epitope of chondroitin sulfate and of carboxy propeptides of type II procollagen for diagnosis of osteochondral fragmentation in horses. Am J Vet Res 1999;60:306309.

    • Search Google Scholar
    • Export Citation
  • 39

    Fuller CJ, Barr AR & Sharif M, et al. Cross-sectional comparison of synovial fluid biochemical markers in equine osteoarthritis and the correlation of these markers with articular cartilage damage. Osteoarthritis Cartilage 2001;9:4955.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 40

    Misumi K, Vilim V & Clegg PD, et al. Measurement of cartilage oligomeric matrix protein (COMP) in normal and affected equine synovial fluids. Osteoarthritis Cartilage 2001;9:119127.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 41

    Muttini A, Petrezzi L & Tinti A, et al. Synovial fluid parameters in normal and osteochondritic hocks of horses with open physis. Boll Soc Ital Biol Sper 1994;70:337344.

    • Search Google Scholar
    • Export Citation
  • 42

    Kawcak CE. Models of equine joint disease. In: Ross MW, Dyson SJ, eds. Diagnosis and management of lameness in the horse. Philadelphia: WB Saunders Co, 2003;594598.

    • Search Google Scholar
    • Export Citation

Advertisement

Use of infrared spectroscopy for diagnosis of traumatic arthritis in horses

Monchanok VijarnsornDepartment of Health Management, University of Prince Edward Island, PE C0A 1T0, Canada

Search for other papers by Monchanok Vijarnsorn in
Current site
Google Scholar
PubMed
Close
 DVM, MSc
,
Christopher B. RileyDepartment of Health Management, University of Prince Edward Island, PE C0A 1T0, Canada

Search for other papers by Christopher B. Riley in
Current site
Google Scholar
PubMed
Close
 BVSc, PhD
,
R. Anthony ShawInstitute for Biodiagnostics, National Research Council of Canada, 435 Ellis Ave, Winnipeg, MB R3B 1Y6, Canada

Search for other papers by R. Anthony Shaw in
Current site
Google Scholar
PubMed
Close
 PhD
,
C. Wayne McIlwraithOrthopaedic Research Center, Department of Clinical Sciences, College of Veterinary Medicine and Biomedical Sciences, Colorado State University, Fort Collins, CO 80523-1678

Search for other papers by C. Wayne McIlwraith in
Current site
Google Scholar
PubMed
Close
 BVSc, PhD
,
Daniel A. J. RyanDepartment of Mathematics and Statistics, Faculty of Science, University of Prince Edward Island, PE C0A 1T0, Canada

Search for other papers by Daniel A. J. Ryan in
Current site
Google Scholar
PubMed
Close
 PhD
,
Patricia L. RoseDepartment Companion Animals, Atlantic Veterinary College, University of Prince Edward Island, PE C0A 1T0, Canada

Search for other papers by Patricia L. Rose in
Current site
Google Scholar
PubMed
Close
 DVM, MS
, and
Elizabeth SpanglerDepartment of Health Management, University of Prince Edward Island, PE C0A 1T0, Canada

Search for other papers by Elizabeth Spangler in
Current site
Google Scholar
PubMed
Close
 DVM, PhD

Abstract

Objective—To evaluate use of infrared spectroscopy for diagnosis of traumatic arthritis in horses.

Animals—48 horses with traumatic arthritis and 5 clinically and radiographically normal horses.

Procedures—Synovial fluid samples were collected from 77 joints in 48 horses with traumatic arthritis. Paired samples (affected and control joints) from 29 horses and independent samples from an affected (n = 12) or control (7) joint from 19 horses were collected for model calibration. A second set of 20 normal validation samples was collected from 5 clinically and radiographically normal horses. Fourier transform infrared spectra of synovial fluids were acquired and manipulated, and data from affected joints were compared with controls to identify spectroscopic features that differed significantly between groups. A classification model that used linear discriminant analysis was developed. Performance of the model was determined by use of the 2 validation datasets.

Results—A classification model based on 3 infrared regions classified spectra from the calibration dataset with overall accuracy of 97% (sensitivity, 93%; specificity, 100%). The model, with cost-adjusted prior probabilities of 0.60:0.40, yielded overall accuracy of 89% (sensitivity, 83%; specificity, 100%) for the first validation sample dataset and 100% correct classification of the second set of independent normal control joints.

Conclusions and Clinical Relevance—The infrared spectroscopic patterns of fluid from joints with traumatic arthritis differed significantly from the corresponding patterns for controls. These alterations in absorption patterns may be used via an appropriate classification algorithm to differentiate the spectra of affected joints from those of controls.

Abstract

Objective—To evaluate use of infrared spectroscopy for diagnosis of traumatic arthritis in horses.

Animals—48 horses with traumatic arthritis and 5 clinically and radiographically normal horses.

Procedures—Synovial fluid samples were collected from 77 joints in 48 horses with traumatic arthritis. Paired samples (affected and control joints) from 29 horses and independent samples from an affected (n = 12) or control (7) joint from 19 horses were collected for model calibration. A second set of 20 normal validation samples was collected from 5 clinically and radiographically normal horses. Fourier transform infrared spectra of synovial fluids were acquired and manipulated, and data from affected joints were compared with controls to identify spectroscopic features that differed significantly between groups. A classification model that used linear discriminant analysis was developed. Performance of the model was determined by use of the 2 validation datasets.

Results—A classification model based on 3 infrared regions classified spectra from the calibration dataset with overall accuracy of 97% (sensitivity, 93%; specificity, 100%). The model, with cost-adjusted prior probabilities of 0.60:0.40, yielded overall accuracy of 89% (sensitivity, 83%; specificity, 100%) for the first validation sample dataset and 100% correct classification of the second set of independent normal control joints.

Conclusions and Clinical Relevance—The infrared spectroscopic patterns of fluid from joints with traumatic arthritis differed significantly from the corresponding patterns for controls. These alterations in absorption patterns may be used via an appropriate classification algorithm to differentiate the spectra of affected joints from those of controls.

Osteoarthritis is a commonly encountered cause of lameness in performance horses and has been implicated as a cause of lameness in 54% of horses.1 Lameness problems have been estimated to result in 68% of days lost in training among racehorses.2 Timely diagnosis and aggressive treatment of traumatically induced osteoarthritis are important to alleviate the effects of inflammation, including pain and reduced function, and are essential to prevent or minimize the development of osteoarthritis.3

Evaluation of joint disease in a horse is facilitated by clinical examination to detect signs of pain and gross anatomical or functional change, evaluate the horse's gait, and localize the problem by use of diagnostic analgesia.3 Other diagnostic aids include radio-graphy, ultrasonography, computed tomography, magnetic resonance imaging, nuclear medical imaging, arthroscopy, and routine synovial fluid analysis.3–5 Although radiography is presently the most practical imaging technique used to aid diagnosis, pathologic changes in articular cartilage cannot be readily assessed, and a lack of sensitivity in the detection of subchondral fragmentation has been reported.6 Nuclear scintigraphy is an advanced diagnostic tool for musculoskeletal disease with high sensitivity but low specificity.7 Factors such as age, breed, and occupation of horses can affect interpretation of the scan.7 Magnetic resonance imaging generates excellent anatomic and pathoanatomic information on articular structures, but the high cost of acquiring and maintaining equipment, the limited availability for use in horses, and the need for general anesthesia for highresolution images have prevented its widespread use.8 None of these tools yield useful biochemical information.

Conventional synovial fluid analyses are not widely used for evaluation of noninfectious joint disease because they rarely provide clinicians with a specific diagnosis.5,9 Recently, ELISA and radioimmunoassaybased evaluations of biomarkers in synovial fluid have been described.9–11 Complex multiple assays are required, individual testing by use of these techniques is expensive, and the relationships of the concentrations of the biomarkers to age, breed, sex, and circadian rhythms are poorly understood.12 Early results suggest promise, but further study is required to determine the clinical usefulness of biomarkers for classifying osteoarthritis.9–12 Presently the means to objectively identify the level of pathologic progression in most cases of traumatic and other forms of osteoarthritis are not available primarily because no generally accepted objective standards exist.6,9 There is a real need for a rapid, economical, practical, and reliable diagnostic test for objective evaluation of joint disease, as well as the unbiased monitoring of responses to treatment.

Infrared spectroscopy is rapidly emerging as a powerful diagnostic probe for biological molecules in humans and other animals.13,14 Infrared spectroscopy measures infrared absorption patterns of molecules when exposed to infrared light.14 An infrared spectrum is obtained when infrared radiation is transmitted through a sample in an FTIR. The fraction of the incident radiation absorbed at a particular wave number (centimeters−1) is determined and displayed as absorption bands on the spectrum.15 These absorption bands correspond to carbon skeletal and functional group vibrations.16 Simple molecules yield simple spectra with well-resolved absorption bands that reflect both structure and concentration.13,17 In a complex sample, compared with a simple sample, the number of functional chemical groups increases, causing the number of absorption bands and the extent of band overlap to increase.17 The infrared spectrum of a biological sample becomes more complex, but the fundamental rule still applies. The infrared spectrum of body fluids or tissues reflects both the structure of the individual infrared active constituents and their relative abundance.14,17 The absorption patterns in the infrared spectra of biological samples may be viewed as biochemical fingerprints that correlate directly with the presence or absence of diseases.14,18 For example, infrared spectroscopy has been used in diagnosis of human diseases such as diabetes mellitus,19 Alzheimer's disease,20 breast tumors,21 and arthritic disorders.22–25 The advantages of an infrared spectroscopic approach in clinical diagnosis are that no reagents are required and automated repetitive analyses can be carried out at very low cost.14 In addition, because the infrared spectrum of biological samples such as synovial fluid reflects the sum of all infrared-active components,26 the infrared spectra of such samples may carry infrared signatures of known and unknown biomarkers, rather than a few novel disease markers.

We hypothesized that traumatic arthritis in horses leads to changes in synovial fluid composition, altering the infrared absorption pattern of synovial fluid samples, and that these spectroscopic changes can be detected and used to differentiate the synovial fluid spectra of joints with traumatic arthritis from the spectra of control samples. The objective of the present study was to evaluate infrared spectroscopy for diagnosis of traumatic arthritis in horses.

Materials and Methods

This study was approved by an animal care committee in accordance with the University of Prince Edward Island policy and the principles outlined in the Guide to the Care and Use of Experimental Animals prepared by the Canadian Council on Animal Care.

Horses and samples—Synovial fluid samples (n = 77) were collected from 48 horses evaluated for arthroscopic removal of osteochondral fragments or intra-articular fracture repair by 1 author (CWM) at 1 center after clinical and radiographic assessments. Samples for model development (calibration) and initial model validation were from racing Thoroughbreds (n = 25) and Quarter Horses (23). These horses were 2 to 5 years old with a median age of 3 years. There were 37 males and 11 females. All horses had clinical evidence of osteochondral fracture of the antebrachiocarpal, midcarpal, or metacarpophalangeal joints (Table 1). Synovial samples were collected aseptically prior to arthroscopic surgery, and samples from contralateral or ipsilateral joints with no evidence of articular fragmentation were also collected as controls.

Table 1—

Anatomic locations and diagnoses in a study of the use of infrared spectroscopy for diagnosis of traumatic arthritis in horses.

JointDiagnosisNo. of synovial fluid samples
Step 1A and 1BStep 2Step 3
AntebrachiocarpalOsteochondral fracture8 (2)10
Ipsilateral-contralateral control8 (6)110
MidcarpalOsteochondral fracture19 (11)10
Ipsilateral-contralateral control19 (8)510
MetacarpophalangealOsteochondral fracture2 (1)40
Condylar fracture of the third metacarpus0 (0)30
Proximal sesamoid fracture0 (0)30
Contralateral control2 (1)10
No. of joints with traumatic arthritis29 (14)120
No. of control joints29 (15)720

Numbers in parentheses indicate number of joints randomly selected from step 1A and used in step 1B.

Step 1A = Infrared region selection. Step 1B = Calibration for classification model. Step 2 = Validation of model with independent within-population samples. Step 3 = Validation of model with independent normal control samples.

Of the 48 horses, paired samples were collected from 29 (1 each from affected and control joints); these were used for calibration of the model. From the remaining 19 horses with traumatic arthritis, independent samples were collected from either an affected (n = 12) or a control (7) joint only; these were used for initial model validation (Table 1). A second set of control synovial fluid samples (n = 20) was collected for further independent validation. These samples were from the left and right antebrachiocarpal and midcarpal joints of 5 horses. Median age of the horses was 4 years, 2 were Trakehner crosses, 3 were Standardbreds, 4 were females, and 1 was male. On the basis of history and results of clinical evaluations, these 5 horses had no evidence of joint disease. A general physical examination and lameness examination were performed by 2 of the authors (CBR and MV). Bilateral radiographs of carpal, metacarpophalangeal, stifle, and tarsal joints were evaluated by 1 author (PLR). Conventional synovial fluid analysis was also performed. All evaluators were unaware of clinical status of the horses. Synovial fluid samples were stored at −80°C in plain cryovials for later batch infrared spectroscopic analysis. Those collected in the United States (CWM) were shipped frozen in a single batch for analysis in Canada.

FTIR—Synovial fluid samples were thawed at 22°C and centrifuged at 2,700 X g for 10 minutes; the supernatants were kept for analyses. Synovial fluid samples were prepared as described27 with the following modification. Briefly, for each sample, an aliquot was drawn and diluted in a KSCNa solution (4 g/L) in a 3:1 synovial fluid–to–KSCN solution. The isolated KSCN absorption peak at approximately 2,060 cm−1 served as a reference band for normalization of the spectral intensities.27,28

Triplicate dry films were made for each sample by applying 8 μL of the diluted synovial fluid preparation evenly in a circular motion onto 5-mm-diameter circular islands on a custom-made, adhesive-masked, silicon microplate; the adhesive mask serves to spatially define and systematically separate the 5-mm islands on the microplate so that sample islands are correctly aligned with the FTIR detector. The synovial films were left to dry at 22°C for 12 hours. After the films were thoroughly dried, the microplate was mounted in a multisamplerb interfaced to the FTIR spectrometer to enable acquisition of infrared spectra. Infrared spectroscopic analyses of all samples were performed during the same period.

Infrared absorbance spectra in the range of 400 to 4,000 cm−1 were recorded with an FTIR spectrometerb equipped with a deuterium tryglycine sulfate detector. For each acquisition, 512 interferograms were signal averaged and Fourier transformedc to generate a spectrum with a nominal resolution of 4 cm−1.27

Data preprocessing and statistical analysis—Triplicate spectra of each sample yielded mean values. By use of spectral manipulation software,d differentiation and smoothing procedures (Savitsky Golay second-order derivatives with second-degree polynomial functions with 15-point smoothing) were performed on all spectra to resolve and enhance weak spectral features and to remove variation in baselines.29

Infrared region selection—The strategy used to find significant (P < 0.01) differences between affected and control joints was to examine the spectroscopic differences for horses that provided paired samples, 1 for an affected joint and the other for a contralateral or an ipsilateral control. Twenty-nine horses yielded such paired synovial fluid samples, resulting in 58 averaged spectra. The pairwise spectroscopic differences in the corresponding spectra (control minus affected) were evaluated in the infrared range of 400 to 1,800 cm−1 (molecular fingerprint region).15 The next step was to seek those subregions where the difference (control minus affected) was significant (P < 0.01). These subregions were identified by use of paired t tests performed with statistical software.e The regions that had significant differences between affected and control samples were identified, and the spectral intensities in each region were then averaged. The averaged value of each of the selected regions was then considered as a variable for inclusion in a classification model.30

Development and calibration of the classification model—To avoid violation of assumptions of independence necessary for discriminant analysis, the 29 horses described previously were randomly assigned into group 1 (n = 15) and group 2 (14). For group 1 (control group), only the spectra from the control joints (n = 15) were used. For group 2 (affected group), only the spectra from the joints with osteochondral fracture (n = 14) were used. This set of 29 spectra provided the basis to calibrate the classification model (Table 1). By use of the set of averaged regional intensities as input variables for each case, stepwise discriminant analysis was then performed by use of proprietary statistical softwaref to select the subset of variables that most contributed to the power of the discriminatory function.31 That subset of variables was then subjected to LDA to find the discriminatory function and rule that best separated the 2 groups (affected vs control), by use of statistical software.g Two sets of cost-adjusted prior probabilities of group membership31 (affected to control ratio, 0.60:0.40 or 0.50:0.50) were selected for this preliminary classification. The posterior probabilities of group membership were calculated for each spectrum. The membership of the spectrum was thus predicted, and the spectrum was assigned to the affected or control group on the basis of its posterior probability. A classification table then revealed the correct classifications for the 29 randomly selected spectra composing the calibration sample set.14,18,32

Validation of the model by use of within-population samples—The remaining 19 of 48 horses that were not used to calibrate the model yielded 19 spectra (7 control and 12 affected spectra) for use as a validation dataset to test the predictive accuracy of the classification model. The classification success rate for this set of spectra was determined and compared with the results for the calibration set.14,18,32

Validation of the model by use of independent normal control samples—The second independent set of samples from 5 control horses that yielded 20 averaged spectra from bilateral antebrachiocarpal and midcarpal joints was used to further characterize the predictive accuracy of the classification model. The classification success rate for this set of spectra was determined and compared with the results for the calibration set (Table 1).

Results

Paired t tests revealed 24 spectral regions in the 400 to 1,800 cm−1 wave number range that had significant differences between the affected and control synovial fluid spectra. From this set of regions, stepwise discriminant procedures resulted in the final selection of 3 regions that most contributed to the discriminatory power of the classification algorithm. These encompassed the wave number ranges 1,245 to 1,257 cm−1, 1,681 to 1,684 cm−1, and 1,691 to 1,694 cm−1(Figure 1).

Figure 1—
Figure 1—

Averaged spectrum (second derivative; dotted line) of a control group of synovial fluid samples and corresponding averaged difference spectrum (control minus affected [samples from horses with traumatic arthritis] spectra; solid line) in a study of use of infrared spectroscopy for diagnosis of traumatic arthritis in horses. Shaded areas (1,245 to 1,257 cm−1, 1,681 to 1,684 cm−1, and 1,691 to 1,694 cm−1) represent the optimal infrared regions for diagnostic classification of the spectra. Negative features in the second-derivative spectrum of the control group correspond to positive features (absorptions) in the original spectrum.

Citation: American Journal of Veterinary Research 67, 8; 10.2460/ajvr.67.8.1286

The classification model developed by use of LDA, with these 3 regional intensities as input for each of the 29 calibration samples, correctly classified 28 of the 29 calibration spectra (Table 2), yielding an overall accuracy of 97%, specificity of 100%, and sensitivity of 93%. Both sets of cost-adjusted prior probabilities give the same classification result.

Table 2—

Calibration dataset from a study of the use of infrared spectroscopy for diagnosis of traumatic arthritis in horses.

Clinical diagnosisInfrared spectroscopy-based diagnosis
ControlOsteochondral fractureTotal
Control15*015
Osteochondral fracture113*14
Total161329

Data are numbers of spectra classified into control and osteo-chondral fracture categories for both sets of cost-adjusted prior probabilities.

Indicates number of spectra correctly classified.

When the classification algorithm was applied to the within-population validation set (n = 19), the LDA classifier with cost-adjusted prior probabilities of 0.60:0.40 (affected-to-control) achieved an overall accuracy of 89%, with 100% specificity and 83% sensitivity (Table 3). With equal prior probabilities of group membership (0.50:0.50), the overall accuracy decreased to 79% (specificity, 100%; sensitivity, 67%). All of the normal control samples that composed the second validation set were classified correctly by use of both sets of cost-adjusted prior probabilities.

Table 3—

Validation dataset from a study of the use of infrared spectroscopy for diagnosis of traumatic arthritis in horses.

Clinical diagnosisInfrared spectroscopy-based diagnosis
ControlOsteochondral fractureTotal
Control7(7)*0(0)7(7)
Osteochondral fracture2(4)10(8)*12(12)
Total9(11)10(8)19(19)

Data are numbers of spectra classified into control and osteo-chondral fracture categories when setting cost-adjusted prior (affected to control ratio) at 0.60:0.40. Numbers in parentheses indicate results when setting equal cost-adjusted prior.

Indicates number of spectra correctly classified.

The basis for these classifications was depicted (Figure 2). With each measured spectrum represented by the triplet of averaged intensities in the 3 subregions that provided optimal classification accuracy, the scatterplot graphically illustrated the separation of affected from control (clustering) spectra for the calibration and both validation datasets.

Figure 2—
Figure 2—

Three-dimensional representation of spectral datasets preprocessed for classification and their division into calibration and validation sets in a study of use of infrared spectroscopy for diagnosis of traumatic arthritis in horses. Each observation is represented by the triplet of intensities in the regions 1,245 to 1,257 cm−1 (infrared region I), 1,681 to 1,684 cm−1 (infrared region II), and 1,691 to 1,694 cm−1 (infrared region III) for the second-derivative spectrum. Data are control spectra (open circles) and osteochodral fracture spectra (closed circles) in calibration set, control spectra (open triangles) and osteochondral fracture spectra (closed triangles) in validation set I, and normal control spectra in validation set II (diamonds).

Citation: American Journal of Veterinary Research 67, 8; 10.2460/ajvr.67.8.1286

Discussion

In this study, significant differences in the infrared absorption pattern of synovial fluid samples were detected for comparison of samples from joints with traumatic arthritis to control samples. The infrared spectra successfully served as biochemical fingerprints to permit diagnosis of traumatic arthritis by means of LDA classification of the processed data, which supported our hypothesis.

The ultimate goal of this type of research is to develop a novel test that aids clinical, and perhaps preclinical, diagnosis of joint disease in horses. In agreement with published recommendations for the development of a new diagnostic test, an exploratory phase was conducted by the authors in a limited number of horses to determine its feasibility and accuracy as a first step.33 The authors chose naturally occurring traumatic arthritis as a model to determine the feasibility and accuracy of the infrared spectroscopic technique. It was thought that if this methodology was determined to be incapable of detecting more severe forms of equine joint disease, its future use to develop a test for subclinical or mild joint disease would be limited. Intra-articular fracture, one of the subtypes of traumat ic arthritis entities in horses, is often preceded by subchondral bone changes and may lead to osteoarthritis if diagnosis and treatment are not prescribed in a timely and appropriate fashion.3,34

Although the features distinguishing the infrared spectra of synovial fluid samples of affected from control joints are not readily interpreted, infrared spectra of biological specimens reflect chemical composition and conformation, as well as possible intermolecular interactions.14,16,35 Articular cartilage damage associated with articular fracture or other types of joint injury induces biochemical changes in the affected joints.3,10 The release of wear-and-tear particles as well as articular cartilage-breakdown products activates synovial cells and chondrocytes to increase the production of cytokines, metalloproteinase enzymes, and inflammatory mediators or other biomarkers that can lead to further damage of the cartilage and joint inflammation.3,10,36–40 It is possible that the infrared changes detected correlated to 1 or all of these molecules; however, because of the complex mixture of organic molecules in synovial fluid, this has not yet been determined. Further study is required to establish the precise linkages, if any exist, between infrared spectra and biomarkers of osteoarthritis.

One other veterinary study41 used a limited infrared spectroscopic technique for evaluation of synovial fluid in 14 clinically normal horses and 2 horses with osteochondrosis. Gross, visually apparent differences between spectra of normal and osteochondritic joints were reported, but a multivariate classification algorithm was not developed. In particular, differences in the relative intensity of the infrared absorption bands at 1,035 and 1,115 cm−1 were detected and suggested to be useful in differentiation between normal and osteochondritic joints. However, both the laboratory technique and the etiology of disease differ from the present study, and it is now well established that a larger number of samples must be examined in each class to derive diagnostic tests applicable to a larger population. Infrared spectroscopy of dry synovial fluid films has also been used in the diagnosis of human arthritis, with LDA classifications based on an optimally selected set of 15 infrared regions between 2,800 and 3,050 cm−1.23 The fact that different spectral regions are required for the present study may be attributable to the differences both in species and the nature of the arthritic conditions examined.

The cost-adjusted prior probabilities represent the group prior probability that a spectrum belongs to 1 of the 2 study groups adjusted for the cost of misclassification.31 Henceforth this will be referred to as the cost-adjusted prior. Because there is no evidence suggesting true prevalence of traumatic arthritis in the study population or guidelines suggesting the exact cost of misclassification, in the present study the costadjusted priors of affected horses compared with control horses were explored on the basis of 2 sets of values, 0.50:0.50 (equal) and 0.60:0.40. If the costadjusted prior is set to be equal, the cost of misclassification and the group prior probability are assumed to be equal. Ability of the test to detect traumatic arthritis (test sensitivity) was lower in the validation sets, compared with calibration sets, when cost-adjusted prior was set to equal. This weakness may be improved by choosing an appropriate cost-adjusted prior. In screening for traumatic arthritis in horses, the aim is to identify as many affected horses as possible. The misclassification cost for failure of the clinician to identify affected horses would be delayed treatment, prolonged recovery period, or an unfavorable treatment outcome because of delayed diagnosis. If an intra-articular fracture is not diagnosed early enough, or treatment is not started early enough, it may lead to osteoarthritis. The implications of false positives are clinically less serious because follow-up diagnostic methods (eg, radiography and arthroscopy) will subsequently detect the false positives. In the present study, the authors suggest setting cost-adjusted priors in favor of disease diagnosis because of the unequal cost of misclassification, as described. At this preliminary stage, the authors presently favor setting the cost-adjusted priors of affected-to-control ratio at 0.60:0.40, which implies a cost that is 1.5 times as great for classifying a horse with traumatic arthritis as healthy relative to classifying a healthy horse as having traumatic arthritis.

For the earlier diagnostic study of human arthritis that used infrared spectroscopy of synovial fluid, the spectral classification method was developed by combining an optimal region selection algorithm with LDA classification.23 The differentiation of infrared spectra of joint fluid from 12 nonarthritic and 74 arthritic patients was achieved, and subsequently, the subclassification of 3 categories of human conditions (rheumatoid arthritis, osteoarthritis, and spondyloarthropathy) were detected with an overall specificity and sensitivity of 100% and 96.5%, respectively. The classification success rates were therefore comparable to those achieved for the present study, despite the differences in species and disease etiology, suggesting that infrared spectroscopy may be generally useful to accurately diagnose a variety of joint diseases in a broad spectrum of species.

The classification accuracy for the within-population validation set of spectra was marginally lower than that for the calibration set when a 0.60:0.40 cost-adjusted prior was used. However, the accuracy and sensitivity of the algorithm for the validation set (79% and 67%, respectively) were considerably reduced when a 0.50:0.50 cost-adjusted prior was used. Although the specificity remained at 100% for either ratio of priors and in all 3 data sets, the sensitivity appeared to vary according to cost-adjusted prior values. The authors of the present study found that cost-adjusted prior values were helpful tools in optimizing the classification of spectra from populations of limited size. An objective estimation of cost-adjusted priors may not be feasible for the equine population at large. The authors suggest that spectral and prevalence data from a larger sample size may be more useful for the future development and optimization of the specificity and sensitivity of this test for clinical use.31,33 This would better enable the scope of spectral variation in the affected population at large to be encompassed and reduce the reliance of the test on estimating cost-adjusted prior values, possibly increasing the number of discriminatory variables for inclusion in the final classification model and thus increasing sensitivity and accuracy.

The spectroscopic data from the 3 significant spectral regions found in our proof-of-concept study yielded robust results in classifying control spectra but was less accurate in classifying affected spectra. This was not unexpected given the number of samples available for this study, which allowed for inclusion of only 3 of the 24 significant variables in the classification model (on the basis of infrared region and stepwise selection procedure). These 3 variables were selected statistically and were the important variables that most contributed to the power of discrimination.31 The discriminatory function and rule based on these selective variables were considered sufficient for accurate classification between groups in the calibration set but clearly did not encompass all possible variations of the affected population. This was indicated by the lower sensitivity (67% to 83%) of the validation set when classifying affected spectra on the basis of 3 significant variables. It is expected that the differences in performance between the calibration and the validation sets will be reduced as more samples become available for analysis. With a larger number of samples in the calibration set, the various degrees of articular cartilage changes and other changes associated with traumatic injuries in the population of affected horses will be better represented. Similarly, the larger the number of samples in the validation set, the more confidence we can have in the discriminatory algorithm's ability to discriminate spectra correctly.32 These preliminary results do address our initial objective and favor the further development of this method of joint disease diagnosis in horses.

In the present study, misclassification of certain disease spectra as controls may have been attributable- to variation of the degree and duration of inflammation among traumatized joints. Possibly, a sample from a joint with mild arthritis was difficult to differentiate from controls because of the limited number of spectral regions in the present classification model. Such variation in the degree of severity is inevitable when studying joint disease in a naturally occurring setting and may contribute substantially to the variability of synovial fluid variables of affected horses, as has been reported in an arthritis biomarker study10 in horses. Other models of osteoarthritis such as osteochondral fragment and forced exercise models may provide more control of the degree of inflammation in the affected group.42 Nonetheless, the ability of the approach in this report to correctly identify control or normal joints (test specificity) in both validation sets clearly revealed the diagnostic potential for this classification algorithm in horses.

ABBREVIATIONS

FTIR

Fourier transform infrared

KSCN

Aqueous potassium thiocyanate

LDA

Linear discriminant analysis

a.

Potassium thiocyanate, SigmaUltra, Sigma-Aldrich Inc, St Louis, Mo.

b.

HTS-XT, Bruker Optics, Milton, ON, Canada.

c.

Opus 4.2, Bruker Optik GmbH, Ettlingen, Germany.

d.

GRAMS/AI 7.02, Thermo Galactic, Salem, NH.

e.

SAS, version 8.02, SAS Institute Inc, Cary, NC.

f.

PROC STEPDISC, version 8.02, SAS Institute Inc, Cary, NC.

g.

PROC DISCRIM, version 8.02, SAS Institute Inc, Cary, NC.

References

  • 1

    Kidd JA, Fuller C, Barr ARS. Osteoarthritis in the horse. Equine Vet Educ 2001;13:160168.

  • 2

    Rossdale PD, Hopes R & Wingfield Digby NJ, et al. Epidemiological study of wastage among racehorses 1982 and 1983. Vet Rec 1985;116:6669.

  • 3

    McIlwraith CW. Diseases of joints, tendons, ligaments, and related structures. In: Stashak TS, ed. Adams' lameness in horses. 5th ed. Philadelphia: Lippincott Williams & Wilkins, 2002;459644.

    • Search Google Scholar
    • Export Citation
  • 4

    Park RD, Wrigley RH, Steyn PF. Equine diagnostic imaging. In: Stashak TS, ed. Adams' lameness in horses. 5th ed. Philadelphia: Lippincott Williams & Wilkins, 2002;185375.

    • Search Google Scholar
    • Export Citation
  • 5

    Van Pelt RW. Interpretation of synovial fluid findings in the horse. J Am Vet Med Assoc 1974;165:9195.

  • 6

    Kawcak CE, McIlwraith CW & Norrdin RW, et al. Clinical effects of exercise on subchondral bone of carpal and metacarpophalangeal joints in horses. Am J Vet Res 2000;61:12521258.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7

    Twardock AR. Equine bone scintigraphic uptake patterns related to age, breed, and occupation. Vet Clin North Am Equine Pract 2001;17:7594.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 8

    Kraft SL, Gavin P. Physical principle and technical considerations for equine computed tomography and magnetic resonance imaging. Vet Clin North Am Equine Pract 2001;17:115130.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9

    McIlwraith CW, Billinghurst RC, Frisbie DD. Current and future diagnostic means to better characterize osteoarthritis in the horse—routine synovial fluid analysis and synovial fluid and serum markers, in Proceedings. 47th American Association Equine Practice Annu Conv 2001;47:171179.

    • Search Google Scholar
    • Export Citation
  • 10

    Bertone AL, Palmer JL, Jones J. Synovial fluid cytokines and eicosanoids as markers of joint disease in horses. Vet Surg 2001;30:528538.

  • 11

    Trumble TN, Trotter GW & Oxford JR, et al. Synovial fluid gelatinase concentrations and matrix metalloproteinase and cytokine expression in naturally occurring joint disease in horses. Am J Vet Res 2001;62:14671477.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12

    Poole AR. Biochemical/immunochemical biomarkers of osteoarthritis: utility for prediction of incident or progressive osteoarthritis. Rheum Dis Clin North Am 2003;29:803818.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 13

    Shaw RA, Mantsch HH. Vibrational biospectroscopy: from plants to animals to humans. A historical perspective. J Mol Structure 1999;480/481:113.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 14

    Shaw RA, Mantsch HH. Infrared spectroscopy in clinical and diagnostic analysis. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;83102.

    • Search Google Scholar
    • Export Citation
  • 15

    Stuart B. Introduction. In: Infrared spectroscopy: fundamentals and applications. Chichester, England: John Wiley & Sons Ltd, 2004;113.

    • Search Google Scholar
    • Export Citation
  • 16

    Coates J. Interpretation of infrared spectra, a practical approach. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;1081510837.

    • Search Google Scholar
    • Export Citation
  • 17

    Dubois J, Shaw RA. IR spectroscopy in clinical and diagnostic applications. Anal Chem 2004;76:361A367A.

  • 18

    Jackson M, Mantsch HH. Infrared spectroscopy, ex vivo tissue analysis by. In: Meyers RA, ed. Encyclopedia of analytical chemistry: applications, theory, and instrumentation. Chichester, England: John Wiley & Sons Ltd, 2000;131156.

    • Search Google Scholar
    • Export Citation
  • 19

    Petrich W, Staib A & Otto M, et al. Correlation between the state of health of blood donors and the corresponding mid-infrared spectra of the serum. Vib Spectrosc 2002;28:117129.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 20

    Pizzi N, Choo LP & Mansfield J, et al. Neural network classification of infrared spectra of control and Alzheimer's affected tissue. Artif Intell Med 1995;7:6779.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21

    Jackson M, Mansfield JR & Dolenko B, et al. Classification of breast tumors by grade and steroid receptor status using pattern recognition analysis of infrared spectra. Cancer Detect Prev 1999;23:245253.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 22

    Canvin JM, Bernatsky S & Hitchon CA, et al. Infrared spectroscopy: shedding light on synovitis in patients with rheumatoid arthritis. Rheumatology (Oxford) 2003;42:7682.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 23

    Eysel HH, Jackson M & Nikulin A, et al. A novel diagnostic test for arthritis: multivariate analysis of infrared spectra of synovial fluid. Biospectroscopy 1997;3:161167.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24

    Shaw RA, Kotowich S & Eysel HH, et al. Arthritis diagnosis based upon the near-infrared spectrum of synovial fluid. Rheumatol Int 1995;15:159165.

  • 25

    Staib A, Dolenko B & Fink DJ, et al. Disease pattern recognition testing for rheumatoid arthritis using infrared spectra of human serum. Clin Chim Acta 2001;308:7989.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26

    Jackson M, Sowa MG, Mantsch HH. Infrared spectroscopy: a new frontier in medicine. Biophys Chem 1997;68:109125.

  • 27

    Shaw RA, Mantsch HH. Multianalyte serum assays from mid-IR spectra of dry films on glass slides. Appl Spectrosc 2000;54:885889.

  • 28

    Low-Ying S, Shaw RA & Leroux M, et al. Quantitation of glucose and urea in whole blood by mid-infrared spectroscopy of dry films. Vib Spectrosc 2002;28:111116.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 29

    Adams MJ. Feature selection and extraction. In: Chemometrics in analytical spectroscopy. Cambridge, England: The Royal Society of Chemistry, 2004;5595.

    • Search Google Scholar
    • Export Citation
  • 30

    Nikulin AE, Dolenko B & Bezabeh T, et al. Near-optimal region selection for feature space reduction: novel preprocessing methods for classifying MR spectra. NMR Biomed 1998;11:209216.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 31

    Khattree R, Naik DN. Discriminant analysis. In: Multivariate data reduction and discrimination with SAS software. Cary, NC: SAS Institute Inc, 2000;211345.

    • Search Google Scholar
    • Export Citation
  • 32

    Adams MJ. Pattern recognition II: supervised learning. In: Chemometrics in analytical spectroscopy. Cambridge, England: The Royal Society of Chemistry, 2004;129160.

    • Search Google Scholar
    • Export Citation
  • 33

    Zhou XH, Obuchowski NA, McClish DK. The design of diagnostic accuracy studies. In: Statistical methods in diagnostic medicine. New York: John Wiley & Sons, 2002;5799.

    • Search Google Scholar
    • Export Citation
  • 34

    Kawcak CE, McIlwraith CW & Norrdin RW, et al. The role of subchondral bone in joint disease: a review. Equine Vet J 2001;33:120126.

  • 35

    Stuart B. Spectral analysis. In: Infrared spectroscopy: fundamentals and applications. Chichester, England: John Wiley & Sons Ltd, 2004;4570.

    • Search Google Scholar
    • Export Citation
  • 36

    Clegg PD, Carter SD. Matrix metalloproteinase-2 and -9 are activated in joint diseases. Equine Vet J 1999;31:324330.

  • 37

    Dimock AN, Siciliano PD, McIlwraith CW. Evidence supporting an increased presence of reactive oxygen species in the affected equine joint. Equine Vet J 2000;32:439443.

    • Search Google Scholar
    • Export Citation
  • 38

    Frisbie DD, Ray CS & Ionescu M, et al. Measurement of synovial fluid and serum concentrations of the 846 epitope of chondroitin sulfate and of carboxy propeptides of type II procollagen for diagnosis of osteochondral fragmentation in horses. Am J Vet Res 1999;60:306309.

    • Search Google Scholar
    • Export Citation
  • 39

    Fuller CJ, Barr AR & Sharif M, et al. Cross-sectional comparison of synovial fluid biochemical markers in equine osteoarthritis and the correlation of these markers with articular cartilage damage. Osteoarthritis Cartilage 2001;9:4955.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 40

    Misumi K, Vilim V & Clegg PD, et al. Measurement of cartilage oligomeric matrix protein (COMP) in normal and affected equine synovial fluids. Osteoarthritis Cartilage 2001;9:119127.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 41

    Muttini A, Petrezzi L & Tinti A, et al. Synovial fluid parameters in normal and osteochondritic hocks of horses with open physis. Boll Soc Ital Biol Sper 1994;70:337344.

    • Search Google Scholar
    • Export Citation
  • 42

    Kawcak CE. Models of equine joint disease. In: Ross MW, Dyson SJ, eds. Diagnosis and management of lameness in the horse. Philadelphia: WB Saunders Co, 2003;594598.

    • Search Google Scholar
    • Export Citation

Contributor Notes

Supported by grants from the Canadian Foundation for Innovation and the Sir James Dunn Animal Welfare Center.

The authors thank Dr. Marcos Lores for technical assistance.

Address correspondence to Dr. Riley.