<title>Accuracy assessment of noninvasive hematocrit measurement based on partial least squares and NIR reflectance spectroscopy</title>

Abstract

Hematocrit (Hct) is one of the most important parameters to monitor when the patient has large blood loss or blood dilution. The current standard method for measuring hematocrit is off-line and invasive. An accurate, continuous, and noninvasive method of measuring hematocrit is highly desired for physicians to respond rapidly in life-threatening situations. A set of instrumental characterization experiments was performed to assess the effects of spectrometer drift and probe placement on patient's forearm. Several factors were investigated in order to minimize the patient-dependent offset encountered in a previous study. Key factors were corrected and tested on ten randomly selected Caucasian patients undergoing cardiac surgery on cardiopulmonary bypass (CPB). A fiber optic probe was attached to the patient's forearm and NIR spectra in the 575-1100 nm region were continuously collected with a spectrograph during the surgery. Blood samples were collected before, during, and after the patient was on CPB. Hct was determined by taking the average of ten spun capillary measurements for each blood sample and paired with spectrum by time. Partial least-squares (PLS) analysis was applied to establish the relationship between spectral changes and Hct changes. The accuracy and sources of measurement errors were assessed by developing calibration models using modified instrumental conditions. Under modified experimental conditions, the improvement in patient-dependent offset was greater than 30%. The optimal models for multiple patients required 8 factors in the region of 581-1000 nm with average standard error of prediction (SEP) of 2.16 and R 2 of 0.594.