Quantification of cerebral blood flow in adults by contrast-enhanced near-infrared spectroscopy: Validation against MRI

Abstract

The purpose of this study was to assess the accuracy of absolute cerebral blood flow (CBF) measurements obtained by dynamic contrast-enhanced (DCE) near-infrared spectroscopy (NIRS) using indocyanine green as a perfusion contrast agent. For validation, CBF was measured independently using the MRI perfusion method arterial spin labeling (ASL). Data were acquired at two sites and under two flow conditions (normocapnia and hypercapnia). Depth sensitivity was enhanced using time-resolved detection, which was demonstrated in a separate set of experiments using a tourniquet to temporally impede scalp blood flow. A strong correlation between CBF measurements from ASL and DCE-NIRS was observed (slope = 0.99 ± 0.08, y-intercept = -1.7 ± 7.4 mL/100 g/min, and R2 = 0.88). Mean difference between the two techniques was 1.9 mL/100 g/min (95% confidence interval ranged from -15 to 19 mL/100g/min and the mean ASL CBF was 75.4 mL/100 g/min). Error analysis showed that structural information and baseline absorption coefficient were needed for optimal CBF reconstruction with DCE-NIRS. This study demonstrated that DCE-NIRS is sensitive to blood flow in the adult brain and can provide accurate CBF measurements with the appropriate modeling techniques.

Keywords: Arterial spin labeling; brain imaging; brain trauma; cerebral blood flow; near-infrared spectroscopy.

Figures

Figure 1.

Change in the number of photons (top row) and mean time-of-flight (bottom row) plotted as a function of time following an intravenous bolus injection of ICG at time = 0 s. Time courses acquired at three source-detector distances (rSD = 1, 3, and 4 cm) are shown before (black) and after (red) inflating the tourniquet (see Scalp Contamination Experiments protocol in Methods). Each time series was averaged across five subjects, and the grey shadowing represents the standard deviation.

Figure 2.

(a) Average Δt>L,max across four subjects (mean ± SD) plotted against ICG dose. The measurements deviate substantially from the expected linear relationship (red line) determined from the average Δ<t>L,max at the smallest dose (0.02 mg/kg), i.e. the line connecting the origin with Δ<t>L,max at 0.02 mg/kg. (b) Simulated Δ<t>L,max plotted against Δμa,Bmax from a two-layer head model (blue circles). The simulated Δ<t>L,max values was generated using equation (S3) with μao= 0.12 cm−1, μso = 11 cm−1, rSD = 3.2 cm, and dC= 1.4 cm. The reconstructed Δ<t>L,max obtained by Taylor expansion (black line, αB = −0.91, βB = 5.47, and γB = −29.92) agreed well with the simulated Δ<t>L,max, while the reconstructed Δ<t>L,max from a linear model (red line) was substantially lower.

Figure 3.

(a) Structural MR image showing the location of the fiducial markers. CBF images acquired from one subject at (b) normocapnia (mean grey matter CBF = 67 mL/100g/min, PETCO2 = 35.8 mmHg) and (c) hypercapnia (mean grey matter CBF = 112 ml/100g/min, PETCO2 = 43.5 mmHg). Corresponding brain (black) and arterial (red) ICG concentration curves are shown in the bottom row. For illustration purposes, the brain concentration curves were scaled by a factor of 20. Included in (d) and (e) are the derived CBF·R(T) function obtained by deconvolution. The grey shaded interval indicates the 45-second window used for the deconvolution. The CBF estimates from the ICG data were 76 and 106 mL/100g/min at normo- and hypercapnia, respectively.

Figure 4.

Regression plot comparing CBF values from DCE-NIRS and ASL. Dashed lines are 95% confidence intervals, and the solid line is the best-fit from linear regression (slope = 0.99 ± 0.08, intercept = −1.7 ± 7.4 mL/100g/min and R2 = 0.88).

Figure 5.

Bland–Altman plot comparing CBF measurements from DCE-NIRS and ASL. Date were grouped by site as described in Figure 4. Mean difference between the two methods across data from both sites is indicated by the solid black line (1.9 mL/100 g/min), which was bound by a 95% confidence interval of −15 to 19 mL/100 g/min.

Figure 6.

Predicted error in estimated CBF due to the use of erroneous input parameters used to generate the Taylor expansion coefficients. The mean error values and standard deviation across subjects data at baseline (left column) and hypercapnia (right column) are presented.