Accuracy of oxygen saturation and total hemoglobin estimates in the neonatal brain using the semi-infinite slab model for FD-NIRS data analysis

Abstract

Frequency domain near-infrared spectroscopy (FD-NIRS) is a non-invasive method for measuring optical absorption in the brain. Common data analysis procedures for FD-NIRS data assume the head is a semi-infinite, homogenous medium. This assumption introduces bias in estimates of absorption (μa ), scattering ( [Formula: see text]), tissue oxygen saturation (StO2), and total hemoglobin (HbT). Previous works have investigated the accuracy of recovered μa values under this assumption. The purpose of this study was to examine the accuracy of recovered StO2 and HbT values in FD-NIRS measurements of the neonatal brain. We used Monte Carlo methods to compute light propagation through a neonate head model in order to simulate FD-NIRS measurements at 690 nm and 830 nm. We recovered μa , [Formula: see text], StO2, and HbT using common analysis procedures that assume a semi-infinite, homogenous medium and compared the recovered values to simulated values. Additionally, we characterized the effects of curvature via simulations on homogenous spheres of varying radius. Lastly, we investigated the effects of varying amounts of extra-axial fluid. Curvature induced underestimation of μa , [Formula: see text], and HbT, but had minimal effects on StO2. For the morphologically normal neonate head model, the mean absolute percent errors (MAPE) of recovered μa values were 12% and 7% for 690 nm and 830 nm, respectively, when source-detector separation was at least 20 mm. The MAPE for recovered StO2 and HbT were 6% and 9%, respectively. Larger relative errors were observed (∼20-30%), especially as StO2 and HbT deviated from normal values. Excess CSF around the brain caused very large errors in μa , [Formula: see text], and HbT, but had little effect on StO2.

Keywords: (170.3660) Light propagation in tissues; (170.5380) Physiology; (300.0300) Spectroscopy.

Fig. 1

(a) Sample segmentation results. The layers from outer to inner are scalp (blue), skull (purple), cerebrospinal fluid (green), gray matter (yellow), and white matter (red). (b) The optical probe used to simulate data. The blue dot in the center was the source position, while the red dots were detector positions.

Fig. 2

Recovered absorption (a–d) and scattering (e–h) coefficients for data simulated on homogenous spheres of varying radius. The solid and dashed lines show the simulated values for 690 nm and 830 nm, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 3

Recovered S_tO₂ (O’s) and HbT (X’s) values for data simulated on homogenous spheres of varying radius. The solid and dashed lines show the simulated values for S_tO₂ and HbT, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 4

Normalized partial pathlength of light through a neonate head model with 70% S_tO₂ and 60 μM HbT as a function or source-detector distance for 690 nm (a) and 830 nm (b). Modulation frequency was 110 MHz.

Fig. 5

Recovered absorption (a–d) and scattering (e–h) coefficients for data simulated on a neonate head model for varying S_tO₂ and fixed HbT in the brain. The solid and dashed lines show the simulated values for 690 nm and 830 nm, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 6

Recovered S_tO₂ (O’s) and HbT (X’s) values for data simulated on neonate head model for varying S_tO₂ and fixed HbT in the brain. The solid and dashed lines show the simulated values for S_tO₂ and HbT, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 7

Recovered absorption (a–d) and scattering (e–h) coefficients for data simulated on a neonate head model for varying HbT and fixed S_tO₂ in the brain. The solid and dashed lines show the simulated values for 690 nm and 830 nm, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 8

Recovered S_tO₂ (O’s) and HbT (X’s) values for data simulated on neonate head model for varying HbT and fixed S_tO₂ in the brain. The solid and dashed lines show the simulated values for S_tO₂ and HbT, respectively. Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.

Fig. 9

Selected simulations showing the effects of varying modulation frequency. (a) Recovered S_tO₂ for simulations with varying S_tO₂ and fixed HbT in the brain. (b) Recovered HbT for simulations with varying HbT and fixed S_tO₂ in the brain. (c) and (d) recovered μ_a values for 690 and 830 nm, respectively, for simulation of varying HbT and fixed S_tO₂ in the brain. The points were plotted as box plots showing the distribution of recovered values due to Monte Carlo noise. The red horizontal lines indicate the target value that was simulated. Source-detector separtion was 25–40 mm.

Fig. 10

The effects of increased CSF on recovered μ_a (a–d) and S_tO₂/HbT (e–h). Source-detector separation increases with each column from left to right. Modulation frequency was 100 MHz.