2.1.

Study Design and Patients

We recruited six patients (two males, four females; age range 57 to 66 years old, median 64) admitted to the Massachusetts General Hospital NeuroICU for aneurysmal SAH. Patients were enrolled under two protocols approved by the MGH Institutional Review Board. The patient’s legally authorized representative provided written informed consent before any procedure related to the research study started.

The optical probe was positioned and attached on the patient’s scalp so as not to interfere with clinical monitoring equipment, and DCS data were recorded continuously, and synchronously to clinically monitor physiological signals. The timing of DCS data acquisition for the patient cohort ranged from postadmission day 2 to 22. Patients were studied on one to three separate days, with recording sessions lasting between 10 min and 20.5 h, as permitted by clinical care. Only data from recording sessions longer than 60 min were included in the analysis and are reported here.

The clinical and study characteristics of all patients are shown in Table 1.

Table 1

Clinical and study characteristics of all patients. Sex: M, male; F, female; SAH grade at admission; HH, Hunt and Hess grade; F, Fisher grade; ruptured aneurysm site; MCA, middle cerebral artery; PComm, posterior communicating artery; optical monitoring; SD, source–detector separation.

2.2.

Diffuse Correlation Spectroscopy Measurements

2.2.2.

Diffuse correlation spectroscopy probes

We designed and built multidistance optical probes targeted to clinical applications. Specifically, we used two types of probes, as shown in Fig. 1: an early version, relatively heavy and rigid, with DCS fibers at two source–detector separations [Fig. 1(a)]; a second, lighter, and more flexible version for DCS-only measurements at up to four separations, three-dimensional printed in a rubber-like material [Fig. 1(b)]. The probes contain the fiber ends and small prisms to provide the contact between the fiber ends and the scalp. Illumination is performed with a $200\text{-}\mu \mathrm{m}$ multimode fiber. A small diffuser is placed between the illumination fiber end and the prism to enlarge the illumination beam [Fig. 1(c)], and enables using higher incident power while remaining within the ANSI standards for safe skin exposure.⁶⁴ For the detection, we use single-mode $5\text{-}\mu \mathrm{m}$ diameter fibers. Three fibers were located at the long separation, and the signals from the three detectors were averaged. The short source–detector separation varied between 5 and 9 mm, and the long separation between 20 and 28 mm, depending on DCS signal quality and space availability. The exact values for each recording are shown in Table 1.

Fig. 1

DCS probes. (a) First probe version. The probe is made of two parts so that the separation can be optimized based on the signal quality (b). Second DCS-only probe version, with multiple fibers at each detector location. (c) Schematic of the illumination prism with a diffuser enlarging the beam diameter (patent⁶³). (d) First probe placed on a patient and attached with gauze and collodion. (e) Second DCS-only probe attached on a patient. (S, source; D, detector).

2.3.

Auxiliary Recordings

Simultaneously to the DCS data, we recorded the patient’s physiological signals that were part of their clinical monitoring. In relevance to the present study, these included invasive mean arterial blood pressure (MAP), ICP, CPP, and white matter perfusion from a thermal diffusion invasive probe (Hemedex Inc., Cambridge, Massachusetts) in four patients. However, the recording of white matter perfusion had to be discarded in two of these patients due to low-signal quality of the measurements. To align the time series of the DCS data and of the physiological data in postprocessing, we sent a transistor-transistor logic pulse train of varying frequencies to the DCS device and to the clinical neuromonitor (Moberg CNS). As proposed in the literature, we computed the cerebrovascular pressure reactivity index PRx¹⁵ as the moving Pearson correlation coefficient, over a 5-min window, between 10-s time-averaged values of ICP and MAP.

2.4.

Data Processing

In brief, we compute a CBF index at each time point, as well as an autoregulation index BFAx. Then, we derive two curves of autoregulation (flow versus pressure and autoregulation index versus pressure) for each recording. The steps are detailed below.

3.1.

Signal Quality

The DCS monitoring duration for each recording varied between 2 and 20 h (mean 9 h), with mean of 7 h of data kept in subsequent analysis for each recording. An average of 18% of the data was discarded because of motion artifacts, or in one case because of the probe getting detached.

3.2.

Correlation Between rCBF_i and Invasive Perfusion Measure

White matter perfusion ${\mathrm{BF}}_{\mathrm{WM}}$ was measured by an invasive diffusion probe with sufficient data quality in two patients. Figure 2 shows the scatter plots of ${\mathrm{rCBF}}_{\mathrm{i}}$ versus ${\mathrm{BF}}_{\mathrm{WM}}$ for the two patients. In patient 3, the ${\mathrm{BF}}_{\mathrm{WM}}$ measured by the invasive diffusion probe showed good correlation (correlation coefficient $R=0.59$) with the DCS-derived ${\mathrm{rCBF}}_{\mathrm{i}}$. Instead, in patient 2, there was no correlation between ${\mathrm{BF}}_{\mathrm{WM}}$ and ${\mathrm{rCBF}}_{\mathrm{i}}$ ($R=-0.03$). Note, however, that when selecting shorter segments of data in this patient, the correlation between the two signals increased significantly. For instance, when considering only the last 4 h of recording (as opposed to the total 16 h), ${\mathrm{BF}}_{\mathrm{WM}}$ and ${\mathrm{rCBF}}_{\mathrm{i}}$ were significantly correlated with a correlation coefficient of $R=0.62$.

Fig. 2

Scatter plots of the DCS-derived relative CBF index versus white matter perfusion measured with the invasive diffusion sensor in two patients. In patient 2, the data points corresponding to the last 4 h of recording are plotted in black.

3.3.

Autoregulation Curves

The results of CBFi measurements and of autoregulation indices BFAx and PRx for all subjects are shown in Figs. 3Fig. 4Fig. 5Fig. 6Fig. 7–8. For each figure, multiple rows correspond to different recordings. The first column displays the simultaneous ${\mathrm{rCBF}}_{\mathrm{i}}$, ICP, and CPP time traces over up to 6 h. The second column displays the ${\mathrm{rCBF}}_{\mathrm{i}}$ versus MAP curves for that recording (i.e., parts of the static autoregulation curve). The third and fourth columns present the evolution with CPP of the autoregulation indices BFAx and PRx, respectively. In columns 2, 3, and 4, the thick line represents the median value of ${\mathrm{rCBF}}_{\mathrm{i}}$, BFAx, and PRx, respectively, and the colored area shows the interquartile range.

Fig. 3

Autoregulatory curves for patient 1. The first panel displays the CPP, ICP, and ${\mathrm{rCBF}}_{\mathrm{i}}$ time traces over 6 h of recording. The second panel displays the ${\mathrm{rCBF}}_{\mathrm{i}}$ versus MAP curve for that recording. Panel 3 displays the association of the blood flow autoregulation index BFAx with CPP. Panel 4 displays the association of the reactivity index PRx with CPP. In panels 2, 3, and 4, the thick line represents the median value in that MAP (CPP) segment of ${\mathrm{rCBF}}_{\mathrm{i}}$, BFAx, and PRx, respectively, and the colored area shows the interquartile range.

Fig. 4

Time series, autoregulatory curves, and autoregulation indices for patient 2. The two rows correspond to the 2 days of recordings. The panels are the same as for Fig. 3.

Fig. 5

Time series, autoregulatory curves, and autoregulation indices for patient 3. The panels are the same as for Fig. 3.

Fig. 6

Time series, autoregulatory curves, and autoregulation indices for patient 4. The panels are the same as for Fig. 3.

Fig. 7

Time series, autoregulatory curves, and autoregulation indices for patient 5. The panels are the same as for Fig. 3.

Fig. 8

Time series, autoregulatory curves, and autoregulation indices for patient 6. The panels are the same as for Fig. 3.

4.1.

Feasibility of Measurements

The first goal of our study was to assess feasibility of long-term monitoring with DCS in neurocritical care patients. We could monitor patients with good signal quality for up to 16 h continuously. Measurements in this challenging population are complicated by multiple factors: heterogeneity of the pathology and severity across patients; presence of invasive and noninvasive neuromonitoring devices on the scalp limiting the space available for DCS measurements; deterioration of the DCS signal potentially arising from craniotomy, stitches, invasive bolt, edema, or extra-axial blood; and imaging exams or clinical care resulting in displacement of the probe. For these reasons, it was not possible in this pilot study to control for all parameters, and our measurement conditions present high variability across patients, such as day after admission, number and duration of recordings, location of the DCS probe, and exact source–detector separations.

Nonetheless, these first measurements allowed us to demonstrate the feasibility of DCS monitoring in the NeuroICU and to identify areas of improvements. For instance, we developed during the study a new probe, smaller, lighter, and more flexible, and therefore more adequate for this population.

4.2.

Interpretation of rCBF_i and Autoregulation Curves

Due to the limitation of the multidistance frequency-domain method in the presence of a thick superficial layer and the constraints of space on the head, we did not attempt to measure the absolute optical properties of the head with our frequency-domain system.^66–71 Instead, we assumed fixed optical properties and used a semi-infinite homogeneous model for the head anatomy. We, therefore, cannot trust the absolute values of CBF that we retrieve or compare them across subjects. For these reasons, we limited our analysis on an individual patient basis to relative blood flow changes that are independent from tissue optical properties.

The retrieved autoregulation curves that pass our quality criteria are physiologically plausible, showing linear dependency of flow versus pressure at low ($<90\mathrm{mmHg}$) MAP, and reaching a plateau of regulated flow at intermediate pressure (90 to 110 mmHg) in some patients.

The transition between the regions of preserved or impaired autoregulation occurs at a different perfusion pressure for each patient. This high variability in the results is not surprising considering the heterogeneity of the patients and measuring conditions.

Our data analysis relies on the assumption that autoregulation remains unchanged over the whole duration of each recording so that we can group all data points to derive a static autoregulation curve. However, it is possible that this assumption does not hold true, and that the autoregulatory capacity of each patient in this severely injured population varies dynamically over time.¹⁴ As future studies pair these data with clinical conditions or assess the impact of interventions, we note that the minimal duration necessary to achieve an autoregulatory will relate to how dynamic the range in CBF is; a restricted range of CBF will only minimally explore the autoregulatory capacity.

The second underlying assumption in interpreting our results is that we are monitoring the global autoregulatory capacity of the brain. In 4 out of 6 patients, the DCS probe was located on the hemisphere contralateral to the main injury, whereas in two patients, it was placed on the ipsilateral side. Available space around clinical neuromonitoring equipment on the scalp as well as DCS signal quality dictated these choices. As we are looking at a single location with a sensitivity limited to a few cubic centimeters below the probe, it is possible that multiregional or bihemispheric measurements would reveal regional heterogeneity in the autoregulation or metabolism of the brain, often a result of the heterogeneity of vasospasms, epileptiform and seizure activity, or cortical spreading depolarizations following SAH.^72–75 Further studies monitoring different regions of the brain with additional DCS channels would be required to address this issue. Note, however, that, except for EEG, most neuromonitoring modalities only provide a global (ICP) or on the contrary a very local (laser Doppler flowmetry, brain tissue oxygen tension) surrogate of CBF, or at best pauci-regional measure (TCD, cerebral oximeter). We, therefore, believe that DCS has its space in the neuromonitoring arena to complement these other modalities, by providing a noninvasive, multiregional, and continuous measure of blood flow.

4.3.

Validation of rCBF_i and Autoregulation Curves with Other Modalities

Despite the plausible interpretation of the autoregulation curves, the first main limitation of the present study at this stage is that we cannot validate these physiological interpretations as there is no standard measure of CBF or autoregulation in these patients. The closest measure of CBF that was available in two patients was white matter perfusion ${\mathrm{BF}}_{\mathrm{WM}}$ assessed with an invasive diffusion sensor. However, ${\mathrm{BF}}_{\mathrm{WM}}$ measurements are not always reliable as they are very sensitive to placement and require calibration. In one patient, ${\mathrm{BF}}_{\mathrm{WM}}$ and ${\mathrm{rCBF}}_{\mathrm{i}}$ were not correlated when considering the whole monitoring period. The two signals presented better correlation over shorter segments (4 h). It is possible that the low correlation over the whole recording arises from sudden offsets in the signals or from a significant clinical change midrecording. In addition, the two devices measure two different parameters often in different locations. With these caveats in mind, the observed significant correlation between the invasive measure of perfusion and the DCS-derived ${\mathrm{CBF}}_{\mathrm{i}}$, on patient 3 during the whole recordings, and in patient 2 over a few hours, provides some confidence in the noninvasive measure of CBF we obtain with DCS.

4.4.

Extra-Cerebral Contamination

The second major limitation of our study is the unknown magnitude of the extra-cerebral contribution to the DCS signal. The issue is particularly challenging in the NeuroICU population as all patients were severely injured and therefore susceptible to strongly impaired CA. It complicates the interpretation of the lack of autoregulation we observed in some patients: it could arise from partial contamination of the signal by scalp flow, which is not regulated, or it could be a true measure of severely disrupted autoregulation in the brain.

Different methodologies have been proposed to assess the contribution from extracerebral flow, including measurements at different pressures to modulate scalp flow,^44,76 and/or multilayer modeling of the head.⁷⁷ Here, with the limited information available, we opted for a simple estimate of CBF, obtained by subtracting a fixed portion (70%) of the ${\mathrm{BF}}_{\mathrm{i}}$ value at the short separation from that at the long separation. This is a rough approximation, and the method is likely to over- or under-estimate the scalp contribution depending on individual patient’s anatomy and source–detector separation. Further measurements and modeling studies will be required to improve the quantification of CBF and the minimization of extra-cerebral flow.

In parallel, different paths are conceivable in the future to improve brain sensitivity during clinical monitoring: increased source–detector separation with higher source power to maintain reasonable signal-to-noise ratio; optimization of the DCS probe location on an individual basis after observation of the patient’s CT or MRI scan to minimize the distance to the brain; implementation of time-domain DCS, a recent technological innovation with improved depth sensitivity.³⁹

4.5.

Consistency Between rCBF_i and BFAx

We introduced the BFAx index of autoregulation to characterize the correlation between MAP and CBFi fluctuations. It is modeled on other indices defined in the literature, which have shown utility for monitoring CA in SAH patients. For instance, Soehle et al.⁷⁸ showed increased Mx during cerebral vasospasm, demonstrating impaired CA. Jaeger et al.⁷⁹ observed that ORx, an index of brain tissue oxygen pressure reactivity, indicated impaired autoregulation in patients who develop delayed infarction after SAH.

In general, the observed results for the ${\mathrm{rCBF}}_{\mathrm{i}}$ curves and the BFAx curves are consistent, in that a high slope of ${\mathrm{rCBF}}_{\mathrm{i}}$ versus MAP is accompanied by a high reactivity index, illustrative of disrupted autoregulation, whereas regions of mainly constant ${\mathrm{rCBF}}_{\mathrm{i}}$ correspond to low BFAx, suggestive of preserved autoregulation. There are a few exceptions though, notably in patient 5, who exhibits an inverted response in CBF that decreases for increasing CPP while the BFAx curves presents the expected behavior. As mentioned before, it is interesting to note that the inverted response arises from very slow fluctuations (30 to 60 min) while the value of the reactivity index arises from faster LFOs (10 s). The reason for the paradoxical response in ${\mathrm{rCBF}}_{\mathrm{i}}$ at very low frequencies is unclear, but the clinical association of this finding was with a cyclic variation in respiratory rate (recurring cycles of hyperventilation followed by hypoventilation), in which the hyperventilatory portion of the cycle is typically associated with increased arousal (Cheyne–Stoke respiratory pattern) and thus may increase MAP while the hyperventilation would directly lower CBF. In general, a continuous index of autoregulation based on 10 s time-scale oscillations may be more appropriate than the static curve of autoregulation to describe the fluctuations of autoregulatory capacity in these patients.

While both the reactivity index PRx and this newly introduced autoregulation index BFAx should reflect the autoregulatory capacity of the brain, the agreement between the two is only modest in the present study, with only 5 recordings out of 10 displaying consistent behaviors between the two indices. The disagreement between the two indices could arise from a number of reasons. First, it is important to note that they reflect two different cerebral metrics: ICP, which is believed to reflect CBV, and the DCS-derived measure of CBF. Second, while PRx assesses a global vasomotor reactivity,⁸⁰ BFAx reflects a regional vascular autoregulation. Previous studies have noted discrepancies between PRx and the mean index of cerebrovascular autoregulation Mx, derived from TCD measure of CBFv.^15,81 In a cohort of 345 patients,⁸¹ there was only moderate agreement between PRx and Mx, and the difference between the two indices was stronger for high values of ICP (30 mmHg). One explanation proposed by Zweifel et al.¹⁵ is that the brain compliance at low ICP levels could buffer CBV changes, resulting in a low PRx, whereas for ICP rising above 30 mmHg, PRx increases sharply. While ICP was generally high in the patients we measured, it rarely rose above 20 mmHg. In addition, Zweifel et al. noted that despite the differences between PRx and Mx, the two indices nonetheless showed a highly significant correlation of 0.4 to 0.6. We observe larger differences in our data.

In addition to the fact that PRx and BFAx derive from distinct physiological quantities (CBV or vessel diameter, versus CBF), and reflect processes of different spatial extents (global versus regional), other factors can explain the discrepancy between the two indices. For instance, potential delays between CPP and ICP or ${\mathrm{CBF}}_{\mathrm{i}}$ fluctuations will not be captured by the zero-lag correlation indices. Finally, the measure of ICP can sometimes be unreliable due to the difficulty of ventricular drain placement, as observed for instance in patient 3.

Because of these limitations, the BFAx index will require further investigation in larger studies, ideally with a simultaneous invasive measure of cerebral perfusion for validation. It may provide a future avenue for monitoring autoregulation both continuously and noninvasively in patients with less severe conditions.