Differentiation between different types of femoral head tissue by near-infrared spectroscopy and ensemble-based extreme learning machines

Osteonecrosis of the femoral head (ONFH) is one of common orthopedic clinical diseases. The present work explores the feasibility of discriminating normal and osteonecrosis tissues by near-infrared (NIR spectroscopy) and chemometrics. A dataset consisted of 128 samples are prepared for experiment. Principal component analysis (PCA) was used for exploratory analysis. Relief-based feature selection algorithm used for variable compression and an ensemble-based extreme learning machines modeling method (EELM) are proposed to construct the diagnostic model. Partial least squares-discriminant analysis (PLS-DA) was used as the reference. The results showed that the EELM can provide a satisfactory model using only 200 variables and achieve a total accuracy of 95.2 % on the independent test sets, while the corresponding PLS-DA exhibits only 78.5 % accuracy, indicating that NIR spectroscopy combined with the proposed EELM algorithm is a potential tool for discriminating ONFH tissues from normal ones. Such a procedure can also be applied to other similar tasks.