Determination of triglycerides in human plasma using near-infrared spectroscopy and multivariate calibration methods

The quantitative analysis of triglycerides in human blood plasma was studied without previous sample pre-treatment using direct near-infrared spectroscopy absorption measurements. A comparison of several multivariate calibration techniques, such as partial least squares (PLS), multiple linear regression (MLR) with feature selection by a genetic algorithm, and a neural network, was performed. Several pre-processing data techniques were tested and the best one was obtained using Fourier transform to reduce the random noise, followed by water spectrum subtraction and the first derivative. Also, different spectral ranges were tested, and the range 5500-6870cm^-1 was chosen. The three calibration techniques presented similar performance with relative errors around 9%. In this case, we choose the simplest calibration model which is MLR, because it is more easily implemented in routine practice.