Detection of Brain Hypoxia Based on Noninvasive Optical Monitoring of Cerebral Blood Flow with Diffuse Correlation Spectroscopy

Abstract

Background: Diffuse correlation spectroscopy (DCS) noninvasively permits continuous, quantitative, bedside measurements of cerebral blood flow (CBF). To test whether optical monitoring (OM) can detect decrements in CBF producing cerebral hypoxia, we applied the OM technique continuously to probe brain-injured patients who also had invasive brain tissue oxygen (PbO₂) monitors.

Methods: Comatose patients with a Glasgow Coma Score (GCS) < 8) were enrolled in an IRB-approved protocol after obtaining informed consent from the legally authorized representative. Patients underwent 6-8 h of daily monitoring. Brain PbO₂ was measured with a Clark electrode. Absolute CBF was monitored with DCS, calibrated by perfusion measurements based on intravenous indocyanine green bolus administration. Variation of optical CBF and mean arterial pressure (MAP) from baseline was measured during periods of brain hypoxia (defined as a drop in PbO₂ below 19 mmHg for more than 6 min from baseline (PbO₂ > 21 mmHg). In a secondary analysis, we compared optical CBF and MAP during randomly selected 12-min periods of "normal" (> 21 mmHg) and "low" (< 19 mmHg) PbO₂. Receiver operator characteristic (ROC) and logistic regression analysis were employed to assess the utility of optical CBF, MAP, and the two-variable combination, for discrimination of brain hypoxia from normal brain oxygen tension.

Results: Seven patients were enrolled and monitored for a total of 17 days. Baseline-normalized MAP and CBF significantly decreased during brain hypoxia events (p < 0.05). Through use of randomly selected, temporally sparse windows of low and high PbO₂, we observed that both MAP and optical CBF discriminated between periods of brain hypoxia and normal brain oxygen tension (ROC AUC 0.761, 0.762, respectively). Further, combining these variables using logistic regression analysis markedly improved the ability to distinguish low- and high-PbO₂ epochs (AUC 0.876).

Conclusions: The data suggest optical techniques may be able to provide continuous individualized CBF measurement to indicate occurrence of brain hypoxia and guide brain-directed therapy.

Keywords: Brain ischemia; Cerebral blood flow; Cerebral ischemia; Cerebral metabolic rate; Clark electrode; Coma; Diffuse correlation spectroscopy; Hypoxia; Hypoxia neuromonitoring; Indocyanine green; Near-infrared spectroscopy; Neuromonitoring; Oxygen extraction fraction.

Figure 1.

Configuration of optical monitoring probe and intracranial monitors. The optode was placed on the forehead adjacent to the intracranial monitoring bolt.

Figure 2.

Method of ICG-calibrated diffuse correlation spectroscopy (DCS) for absolute, real time CBF measurements. (A) An infrared light source is used to probe turbid media with moving particles (i.e., red blood cells; red disks at time t, light red disks at time t+Δt). Specifically, light propagates diffusively through the tissue along random walk pathways, and is scattered by moving red blood cells. The light scattered back from the tissue is measured at a detector placed adjacent to the source (source-detector separation is 2.5 cm). In our studies, the source/detector combination was incorporated into a single non-invasive optode patch. Light scattering by moving particles induces rapid (i.e., μs) temporal fluctuations in the detected speckle intensity, which are quantitatively characterized by the normalized intensity autocorrelation function. (B) Changes in the decay of the autocorrelation function over time are due to changes in CBF, which allows for a relative CBF index to be calculated over time. The rCBF index can be converted to absolute CBF by concurrent near infrared spectroscopic measurement of the transit of ICG through brain tissue using the same optode.

Figure 3.

Overview of method used for ICG calibrated NIRS for absolute CBF measurement. (A) An intravenous bolus of ICG is given (0.1 mg/kg), and the temporal ICG concentration is measured in the arterial circulation via a custom pulse oximeter (or dye densitometer, with wavelengths of 804 and 938 nm that are optimal for ICG measurement) and in brain tissue via time-resolved DOS (using the optode with a source-detector separation of 3.2 cm, wavelength of 810 nm). The arterial and brain tissue ICG concentrations were sampled at 1 Hz. (B) Example brain tissue ICG and arterial impulse response function from a patient is shown. Tissue ICG concentration is equivalent to the convolution of the arterial impulse response (i.e., the arterial ICG concentration) and CBF×R(t), where R(t) is the fraction of ICG that remains in tissue at time t. Absolute CBF is calculated by deconvolving the tissue ICG concentration with the arterial ICG concentration to obtain CBF×R(t), and then taking the t=0 intercept (since R(0) = 1 by definition).

Figure 4.

Method for identifying hypoxic events. The plot shows ~1 hour of continuous PbO₂ data from subject OM-14. An episode of brain hypoxia was defined as the average PbO₂ in a 6 minute period of sustained PbO₂ < 20 mmHg that followed a 3 minute period where PbO₂ transitioned from > 20 mmHg to < 19 mmHg. The pre-hypoxia baseline was defined as the average PbO₂ in a 6 minute period of time that occurred 6 minutes prior to the transition to brain hypoxia.

Figure 5:

Scatter plot of MAP (mmHg), CBF (ml/100g/min), and SO₂ (fraction) vs. PbO₂ (mmHg) for randomly selected windows 12 minutes long (as discussed in the text). Fit lines and 95% confidence intervals are shown in black. CBF is substantially more sensitive to changes in PbO₂ than either of the other variables.

Figure 6:

(A) Brain tissue oxygen measurements during hypoxic events (B) Mean arterial pressure (MAP) and (C) cerebral blood flow relative to baseline (rF =CBF/CBF_Baseline)] during hypoxic events. Median baseline CBF was 13.9 ml/100gm/min (IQR 8.6:16.5). MAP (p = 0.00035) and rF (p = 0.01) are significantly lower during hypoxic events. *** = p < 0.001, ** = p < 0.01.

Figure 7:

Receiver Operating Characteristic (ROC) curves for MAP and optically measured CBF to separate randomly and sparsely chosen windows of low and high PbO₂. MAP and absolute CBF were associated with low PbO₂ (ROC AUC 0.761, 0.762 respectively). Combining these variables using logistic regression analysis markedly improved the ability to distinguish low and high PbO₂ epochs (AUC 0.876). TPR-True positive rate, FPR-False positive rate.