Optimal Regression Method for Near-Infrared Spectroscopic Evaluation of Articular Cartilage.

摘要: Near-infrared (NIR) spectroscopy has been successful in nondestructive assessment of biological tissue properties, such as stiffness articular cartilage, and is proposed to be used clinical arthroscopies. spectroscopic data include absorbance values from a broad wavelength region resulting large number contributing factors. This spectrum includes information potentially noisy variables, which may contribute errors during regression analysis. We hypothesized that partial least squares (PLSR) an optimal multivariate technique requires application variable selection methods further improve the performance NIR spectroscopy-based prediction cartilage including instantaneous, equilibrium, dynamic moduli thickness. To test this hypothesis, we conducted for first time comparative analysis techniques, included principal component (PCR), PLSR, ridge regression, absolute shrinkage operator (Lasso), version support vector machines (LS-SVM) on spectral equine cartilage. Additionally, evaluated effect methods, Monte Carlo uninformative elimination (MC-UVE), competitive adaptive reweighted sampling (CARS), combination population (VCPA), backward interval PLS (BiPLS), genetic algorithm (GA), jackknife, technique. The PLSR was found tool (R = 75.6%, R 64.9%) data; simplified models enabling use lesser components. However, improvements model with were statistically insignificant. Thus, recommended properties its spectra.