| Abstract | Synovial fluid aspirates have been characterized by measuring their visible/near-infrared spectra (400–2500 nm). The hypothesis tested in this study is that the spectra contain sufficient information to serve as an aid in the diagnosis and/or staging of arthritic disorders. The concentrations of all major constituents are carried implicitly in the spectra, and in this sense this approach is similar in spirit to conventional synovial fluid analysis. The distinguishing feature of this method is that we have not converted the raw data (spectra) explicitly to analytical information. Rather, we have used automated pattern recognition methods to identify significant characteristics of the spectra themselves. A total of 109 spectra were measured and split into three classes according to the disease (osteoarthritis, rheumatoid arthritis, or spondyloarthropathy) affecting the patient from whom the synovial fluid sample was taken. An automated classification method was then trained by correlating features derived from these spectra to the clinical diagnoses. The robustness of the classification was validated using the leave-one-out cross-validation method, i.e., by training on all but one of the spectra and using the resulting model to predict the classification for the spectrum that is left out. The result derived by following this procedure for each of the spectra was that 105 of the 109 predicted classifications correctly matched the clinical diagnosis. These results suggest that the near-infrared spectrum of synovial fluid is sufficient to allow diagnosis of the disease affecting the joint from which the aspirate is drawn. |
|---|
