Standardization Method for Correcting Spectral Differences acrossMultiple Units of a Portable Near Infrared – Based Medical Monitor

摘要

We have previously demonstrated the correlation of continuous-wave near infrared (CW-NIR) tissuemeasurements, to blood and tissue metabolic parameters using Partial Least Squares (PLS) regression. Thepractical use of this non-invasive measurement technique depends on the transfer of PLS calibrationmodels from a single calibration unit to multiple secondary units. Variations in the spectral characteristicsof the optical components across multiple units result in marked differences in the spectral output,preventing the direct transfer of parameter models from one unit to another. Consequently, we havedeveloped a method for standardizing the spectral output across units that utilizes physical, traceable,reference materials for aligning the wavelength and intensity axes to fixed values, followed by spectralnormalization via Standard Normal Variate transformation. The approach employed in this study adjuststhe slope and bias differences in the optical spectra across multiple units, without the loss of usefulinformation needed for parameter estimation. In this study, phantoms containing Agar, intralipid andlyophilized human hemoglobin (met-hemoglobin) were used to mimic human tissue. Using PLS regression,a hemoglobin calibration model was developed on the tissue-like phantoms on a prototype of the portableNIR medical monitor. The calibration model was successfully transferred to a second, distinctly differentsystem. The Root Mean Squared Error of Prediction of met-hemoglobin in the phantom samples measuredin the second system, improved from 4.94g/dl to 1.15g/dl after the standardization procedure. Thiscompares favorably the PLS model error on the primary instrument (0.94g/dl).