Support Vector Regression for Non-invasive Detection of Human Hemoglobin / 分析化学

ABSTRACT

To facilitate noninvasive diagnosis of anemia, high-performance and portable near infrared (NIR) spectrometer for human blood constituents was designed and fabricated based on linear variable filter (LVF).Meanwhile, the performance of support vector regression (SVR) model for quantitative analysis of human hemoglobin (Hb) was investigated.Spectral data were collected noninvasively from 100 volunteers by self-designed LVF-NIR spectrometer, then divided into calibration set, validation set 1 and 2.To establish a robust SVR model, grid search method was applied to optimize the penalty parameter and kernel function parameter c=5.28, g=0.33.Then, Hb levels in the validation 1 and 2 sets were quantitatively analyzed.The results showed that the root mean square error of prediction (RMSEP) were 10.20 g/L and 10.85 g/L, respectively, and the relative RMSEP (R-RMSEP) were 6.85% and 7.48%, respectively.The results indicated that the SVR model had high prediction accuracy to Hb level and adaptability to different samples, and could satisfy the requirements of clinical measurement.Based on the SVR algorithm, the self-designed LVF-NIR spectrometer has a wide application prospect in the field of non-invasive anemia diagnosis.