Blood diagnostics using fiberoptic evanescent wave spectroscopy and neural networks analysis

A spectral analysis of human blood serum was undertaken by fiberoptic evanescent wave spectroscopy (FEWS) using a Fourier transform infrared (FTIR) spectrometer. The blood serum samples were Introduced into a special cell designed for FEWS of liquids, with an IR transmitting silver halide fiber as the sensing element. The spectra were analyzed by models of neural networks (NN), and the concentrations of protein, cholesterol and uric acid in human blood serum were obtained. Our results are in agreement with those obtained by ordinary chemical enzymatic methods and multivariate calibration methods. The estimated prediction errors obtained (in percent of the average value) were 4.7% for total protein, 22% for cholesterol and 35% for uric acid. This method can be used for in-situ real-time blood analysis. (C) 1997 Elsevier Science S.A.