Reagentless Vis-NIR Spectroscopy Point-of-Care for Feline Total White Blood Cell Counts

Abstract

Spectral point-of-care technology is reagentless with minimal sampling (<10 μL) and can be performed in real-time. White blood cells are non-dominant in blood and in spectral information, suffering significant interferences from dominant constituents such as red blood cells, hemoglobin and billirubin. White blood cells of a bigger size can account for 0.5% to 22.5% of blood spectra information. Knowledge expansion was performed using data augmentation through the hybridization of 94 real-world blood samples into 300 synthetic data samples. Synthetic data samples are representative of real-world data, expanding the detailed spectral information through sample hybridization, allowing us to unscramble the spectral white blood cell information from spectra, with correlations of 0.7975 to 0.8397 and a mean absolute error of 32.25% to 34.13%; furthermore, we achieved a diagnostic efficiency between 83% and 100% inside the reference interval (5.5 to 19.5 × 109 cell/L), and 85.11% for cases with extreme high white blood cell counts. At the covariance mode level, white blood cells are quantified using orthogonal information on red blood cells, maximizing sensitivity and specificity towards white blood cells, and avoiding the use of non-specific natural correlations present in the dataset; thus, the specifity of white blood cells spectral information is increased. The presented research is a step towards high-specificity, reagentless, miniaturized spectral point-of-care hematology technology for Veterinary Medicine.

Keywords: artificial intelligence; point-of-care; spectroscopy; white blood cells.

Figure 1

Total white blood cell counts: (a) sampling and spectroscopy procedures—venipuncture performed at the jugular (2) or cephalic (3) veins for hemogram analysis and single drop for POC spectral recording; a single drop can also be collected at the auricular vein (1). (b) Data augmentation through hibridization of real-world data (RWD) into synthetic spectroscopy data (SSD), validation procedures using RWD and SSD as optimization and validation datasets to test SSD representativity of RWD and knowledgbase expansion.

Figure 2

WBC spectral information: (a) cat blood spectra () low WBCs, () high WBC examples, and () RBC Hgb major inteference bands (539–576 nm); (b) PCA scores of hemogram counts; and (c) Cell packing, scattering and WBC absorbance effects and impact on observed spectra; (d) PLS scores of blood spectra—translating maximum covariance to WBCs, where • SSD blood samples, • RWD blood samples, • low WBCs and • high WBCs and → hemogram PCA loadings and main gradient direction in PLS scores space.

Figure 3

WBC prediction for (a) SLAI with SSD as CV optimization and RWD as HO blind test samples; (b) SLAI with RWD and SSD as CV and HO samples, where (•) represent the hybridized SSD samples and (•) the RWD blood samples, respectively. Green shaded rectangle () represents the WBC reference interval for cats (5.5–19.5 × ${10}^9$ cells/L) and red shaded rectangle () represents the ASVCP total allowable error tolerance for high WBC diagnosis.

Figure 4

WBC quantification benchmarks: (a) Pearson correlation coefficient for RWD and SSD, where shaded rectangles represent semi-quantitative () and quantitative () results; (b) MAPE benchmarks where () represents the ASVCP TAE for WBCs (21.45%); and (c) POC WBC diagnostic capacity: () percentage of correct diagnoses within WBC value interval, (– –) accumulated percentage of correct diagnosis, () WBC histogram distibution; and RI for WBC () and () TAE diagnosis error tolerance.

Figure 5

Representative low and high CovMs: (a) PLS scores space where (∘) and (∘) are low and high CovMs sample groups, • SSD blood samples, • RWD blood samples; (b) spectral ROIs used for the quantification of WBCs ( low and high CovM), () and () major and minor RBC/Hgb interference bands; (c,e) high and low CovM sample spectra and corresponding used ROIs to quantify WBCs; (d,f) prediction plot for high and low CovMs.