First-in-clinical application of a time-gated diffuse correlation spectroscopy system at 1064 nm using superconducting nanowire single photon detectors in a neuro intensive care unit

: Recently proposed time-gated diffuse correlation spectroscopy (TG-DCS) has significant advantages compared to conventional continuous wave (CW)-DCS, but it is still in an early stage and clinical capability has yet to be established. The main challenge for TG-DCS is the lower signal-to-noise ratio (SNR) when gating for the deeper traveling late photons. Longer wavelengths, such as 1064 nm have a smaller effective attenuation coefficient and a higher power threshold in humans, which significantly increases the SNR. Here, we demonstrate the clinical utility of TG-DCS at 1064 nm in a case study on a patient with severe traumatic brain injury admitted to the neuro-intensive care unit (neuroICU). We showed a significant correlation between TG-DCS early ( ρ = 0.67) and late ( ρ = 0.76) gated against invasive thermal diffusion flowmetry. We also analyzed TG-DCS at high temporal resolution (50 Hz) to elucidate pulsatile flow data. Overall, this study demonstrates the first clinical translation capability of the TG-DCS system at 1064 nm using a superconducting nanowire single-photon detector.

In the most severe brain injuries, complications due to TBI -termed secondary brain injuries -can occur quickly and cause significant morbidity or mortality. These patients are cared for in neuro-intensive care units (neuroICUs) focused on monitoring the brain for secondary brain injuries. Patients may not be able to report complications and imaging tests are not suitable for the continuous bedside monitoring needed in this setting [2]. Rather, continuous monitoring in most ICUs is limited to measurements of cardiopulmonary function such as cardiac telemetry or arterial blood pressure monitoring. Invasive probes are recommended to measure intracranial pressure (ICP) in selected patients according to the Brain Trauma Foundation Guidelines [1,3]. In some centers, multimodality monitoring (MMM) is used that incorporates not only ICP but also measurements of brain tissue oxygen tension (PbtO2) or regional cerebral blood flow (rCBF). rCBF can be measured from brain tissue directly using a thermal diffusion flowmetry probe (TDF) (Hemedex, Inc; Cambridge, MA, USA) [4]. However, these methods are invasive, require drilling a burr hole through the skull, which increases the risk of hemorrhage and infection. Furthermore, TDF only probes a small volume and suffers from data loss due to automatic recalibration every 30 min.
Laser Doppler flowmetry (LDF) shares a similar issue with TDF in that it measures CBF only from a very small volume of tissue [5]. Transcranial Doppler ultrasound (TCD) is more sensitive to larger vessels, but it is limited by skull thickness, and it is difficult to use for longitudinal monitoring due to the need for a stable orientation of the ultrasound probe [6]. Continuous EEG (cEEG) can detect seizures, periodic discharges, and spreading depolarizations leading to secondary brain ischemia and may contain relevant prognostic information in patients after moderate or severe TBI [7][8][9][10]. Currently, the international neurocritical care community recommends cEEG monitoring in patients after TBI. Characteristic changes in EEG signals have been recognized in response to brain ischemia in correlation with CBF and oxygen metabolism [11,12]. When the CBF is compromised, the metabolic and electrical activity of cortical neurons is affected, leading to changes in the frequency content of EEG [12]. The disadvantage of EEG is its acquisition paradigm, which produces a high-amplitude mixed frequency that makes far-field event detection difficult and the need for interpretation. Furthermore, EEG exhibits a low signal-to-noise ratio (SNR) because of placement on the scalp far from the source of the signal, artifacts common in the ICU setting, and the effects of interventions, such as sedation. Non-invasive EEG therefore cannot always be relied on to inform of evolving secondary brain injury associated with TBI. Thus, there is a need for continuous, non-invasive monitoring of cerebral blood flow at the bedside in the neuroICU setting.
Diffuse optical technologies enable continuous non-invasive measurements of the hemodynamics within the brain microvasculature [6,13]. Near-infrared spectroscopy (NIRS) can determine changes in cerebral oxygenation, but cannot provide a direct measure of CBF [6,13]. Furthermore, oxygen saturation alone may not be sensitive enough to detect secondary brain injury at the later time points of the insult, as oxygen consumption and delivery may reach equilibrium after acute transient alterations [6,13,14]. Significant work has been done on the monitoring of cerebral oxygenation with NIRS [15,16]. For example, Robertson et al. showed a ∼90% sensitivity for the detection of secondary hematomas [17], and Kampfl et al. showed that patients with an ICP of >25 mmHg exhibited significantly reduced NIRS parameters than those with an ICP <25 mmHg [18]. Yet, there are limitations in NIRS monitoring [1,2]. Busch et al. showed that the NIRS approach was limited in detecting cerebral hypoxia in patients with acute brain injury [14]. Leal-Noval et al. found that NIRS was reliably sensitive to detect relatively severe cerebral hypoxia (brain tissue oxygenation (PbtO2) < 12 mmHg), but showed a weak correlation between measurements of invasive PbtO2 and noninvasive tissue oxygenation measured by NIRS [19] Baker indicated that skeletal muscle tissue oxygenation (StO 2 ) could not detect brain hypoxia (AUC = 0.46), but CBF could detect brain hypoxia with a fairly good predictive power (AUC = 0.76) [20]. A recent investigation of frequency-domain NIRS StO 2 monitoring for detection of brain hypoxia also showed variable predictive ability and was not robust across patients and suggested multi-modal monitoring that involves CBF, StO2, cerebral metabolic rate of oxygen (CMRO 2 ) changes [21]. These systems work in continuous-wave (CW) or frequency-domain (FD) mode, but it cannot effectively separate the scalp signal from the deeper brain signal. Regression analysis that utilizes short source-detector separations [6,[22][23][24][25][26][27] can provide improvements for separating superficial vs. deeper brain signals. Another method to separate the superficial signal from the deep signal is by using the time-domain NIRS (TD-NIRS) [28,29]. With this technique, signals that travel deeper are selected therefore increasing the sensitivity to cerebral hemodynamics, however, since it is a NIRS technique it does not provide a direct measure of CBF.
A more recently developed diffuse optical technique, diffuse correlation spectroscopy (DCS), probes speckle fluctuations related to blood flow [6,13,[30][31][32][33]. It is ideal for high-risk populations, as it can be used as a portable, non-invasive, continuous bedside monitor for CBF [6,14,20,[34][35][36][37]. There is a much higher contrast between cerebral blood flow vs. scalp blood flow [37]. Thus, we expect that blood flow measurements may provide additional benefits for TBI monitoring [14,37]. In fact, DCS has been shown to have a higher sensitivity to the brain compared to NIRS [38]. DCS has shown promise for noninvasive, continuous measurements of CBF in animals and humans [33,[39][40][41], and recently, in neuroICU settings [6,14,20,[34][35][36][37]. It has been demonstrated that DCS measurements agree with clinically-established modalities [22,25,35]. However, most DCS operates in the continuous wave (CW) domain. There exist several crucial limitations of CW-DCS regarding (1) the ability to obtain absolute blood flow index (BFi) (2) its sensitivity to CBF due to confounding effects from scalp blood flow, (3) the occupation of limited space on the forehead, (4) its relatively shallow probing volume for detecting changes in CBF, and (5) the low SNR when the source-detector separation (>25 mm) is optimized for detecting changes in CBF [6,14,20,[34][35][36][37]. The work by Baker et al. is highly significant in this aspect since it directly tackles the limitations of the CW-DCS approach by applying a more realistic two-layer (scalp and brain) model in their data analysis to separate scalp and brain signals [26]. The contamination from the scalp signal in CW-DCS can be reduced by regression analysis and using short source-detector optode configurations [6,42,43].
Very recently we demonstrated the clinical feasibility of using a new generation photoncounting detector, superconducting nanowire single-photon detector (SNSPD), for CW-DCS with improved SNR at 785 nm [67]. This detector has also been successfully used in phantoms and in-vivo time-domain DCS studies [54,47,48]. In this paper, we show the clinical feasibility of a TG-DCS by using SNSPD, which enabled us to use a SWIR wavelength at 1064 nm in a patient with severe TBI in a neuroICU setting. We compared our TG-DCS results with the invasive TDF monitor. Additionally, we sought to provide evidence that the sampling frequency of TG-DCS is sufficient to resolve pulsatile blood flow by correlating it with a commercial heart rate monitor.

Study design and patient details
A clinical prototype of the TG-DCS system was built and tested in a patient with TBI admitted to the neuroscience intensive care unit (NSICU) at the University of Cincinnati Medical Center after informed consent from a legally authorized representative as approved by the Institutional Review Board (IRB) of the University of Cincinnati. This case involves a 26-year-old man who was involved in a motorcycle accident. He was helmeted but upon admission to the Emergency Department, he was found to have a depressed level of arousal with a Glasgow Coma Scale score of 7T. Both pupils were reactive to light. A non-contrast head CT demonstrated a left subdural hematoma with a left-to-right midline shift ( Fig. 1(A)). Hypertonic saline was administered, and he was taken urgently to the operating room, where he underwent left decompressive hemicraniectomy and evacuation of the clot. His additional injuries included bilateral pulmonary contusions. Commercially available, FDA-approved, invasive multimodality intracranial monitoring (MMM) probes were placed within non-dominant frontal subcortical white matter using a burr hole created at the Kocher's point approximately between scalp EEG channels F4 and C4 (according to measurements using the International 10-20 system; Fig. 1(D)). One of these probes was a thermal diffusion flowmetry device (Bowman Perfusion Monitor; Hemedex, Inc; Cambridge, MA) placed to directly measure rCBF in conjunction with other monitored modalities, including ICP, PbtO2, cerebral micro-dialysis, and electrocorticography as previously described [68].
On post-trauma day 3, the portable TG-DCS system was brought to the NSICU. The system was placed in the patient's room and a total of 1 hr and 20 min of data were collected along with simultaneously recorded MMM measurements. During the duration of the TG-DCS measurement, three events related to CBF occurred: (1) propofol administration was stopped, (2) a neurological examination was performed, and (3) administration of propofol was resumed as shown in Fig. 2. The neurological examination consists of noxious stimulation to provoke arousal in addition to maneuvers to test cranial nerve reflexes, such as pupillary examination.

Optical technique
A portable prototype TG-DCS was developed and placed in the patient's room (Fig. 3(A)) during the measurement period. The optical imaging probe was placed just behind the scalp EEG electrode F4 and above the F8 electrode. The probe consisted of a 1 mm source fiber (Thorlabs FP1000ERT, 0.5 NA) and a few-mode fiber bundle (4 fibers, Corning SMF-28) at 15 mm source-detector separation. The fibers were mounted onto a 5 mm right-angled prism and housed in a custom printed flexible 3D probe (Fig. 3(B)). The probe was ergonomically designed to allow for easy placement and designed to minimize the area of scalp needed, with a width just slightly larger than a US quarter dollar (Fig. 3(C)). The probe was sanitized with an alcohol swab and held in position with a Tegaderm film (3M; Saint Paul, MN). All four channels were connected to the SNSPD to collect the reflected photons at 1064 nm. The SNSPD outputs were connected to the 4-channel time-correlated single-photon counter (TCSPC) (HydraHarp400, Picoquant Inc, Berlin, Germany). The laser source consists of a 1064 nm seed laser, operating at a repetition rate of 80 MHz (QC2D106P-64D0, QDLaser Inc, Kanagawa, Japan). The seed laser was amplified with a 1064 nm optical amplifier (Mantaray-Amp-1064, Cybel LLC, Bethlehem, PA) and was fiber-coupled to a 1 mm multimode fiber that was mounted onto the 3D printed probe. The beam on the face of the prism was expanded to an approximately 4.5 mm diameter spot with a diffuser film and the average power was set below ANSI standards at 150 mW (∼943 mW/cm 2 ). Each channel was sampled at approximately 1 × 10 7 cps.

Data analysis
Time-tagged data was collected in 8 × 10-minute bins, representing a total of ∼140 GB of data at ∼1.75 GB/minute of monitoring. The data was loaded into a custom MATLAB routine on a supercomputer node with 1.5 TB of RAM [69]. The instrument response function (IRF) was obtained by placing a thin scattering layer (Teflon Tape, 3M, Saint Paul, MN) between the source fiber (Thorlabs FP1000ERT, 0.5 NA) and the detector fibers for an acquisition time of 30s [60,53]. The full-width half-maximum (FWHM) of the IRF was measured to be ∼220ps (Fig. 4(A)). Two gates were selected: early gate (EG) (-100 ps from the peak with a width of 10 ps) and late gate (LG) (+650 ps from the peak with a width of 800 ps). Here, to reduce the variations caused by varying optical properties, the TPSF for each channel were summed and the gate locations were chosen relative to the peak of the summed TPSF [57] and fixed for the rest of the analysis (Fig. 4(B)). In this clinical setup, we obtained a beta-value of ∼0.075 based on the laser and detection system, where we used a larger core, few-mode fiber (9 um diameter) to improve SNR while using a single-mode fiber would increase the beta-value with the cost of reduced SNR. By selecting photons within specific gates, the normalized intensity autocorrelation function (g 2 ) in selected gates could be obtained [57]. The EG and LG data were autocorrelated from 5e-7 s to 1e-5 s and integrated for 5 s. [44,49,54] (Fig. 4(C)). The optical properties of µs': 0.83 mm −1 and µa: 0.017 mm −1 [54,47] were used in the g 2 model in the time domain [46,53] to fit the data for the absolute blood flow index (BFi) (Fig. 4(D)). Briefly, the IRF was convoluted with Fig. 2. Data from the invasive multimodal monitor probes (From top to bottom: end-tidal carbon dioxide (EtCO2), heart rate (HR), oxygen saturation (SpO2), temperature from the TDF probe (Tperf), perfusion from the TDF probe, mean arterial blood pressure (ABP, mean), cerebral perfusion pressure (CPP), intracranial pressure (ICP), pressure reactivity index (PRx)) are plotted with the blood flow index (BFi) from the early-gated time-gated DCS (TG-DCS-EG) (-100 ps from the peak, with a width of 10 ps) and late-gated (TG-DCS-LG) (650 ps from the peak with 800 ps width) blood flow index. The 3 green line represents (1) the time that sedation is stopped, 2) the neurological examination is performed, and (3) the sedation is restarted. the analytical solution to the time-domain g 2 with the assumed optical properties [46,53]. The corrected g 2 curve is fitted to the experimental g 2 curve to obtain the BFi. To increase the SNR, the experimental g 2 curve from the four channels were averaged before it was fitted.  LG) gates highlighted. The EG opens -100ps from the peak with a width of 10ps (dashed line), while the LG opens +650ps from the peak with a width of 800ps (red overlay) (C) Representative g 2 curve from the LG patient data that is fitted with the time-gated g 2 model. (D) The absolute LG BFi obtained from the fit.

Results and discussion
We obtained a Spearman correlation coefficient (ρ) between TD-DCS BFi and TDF rCBF of 0.67 (EG) and 0.76 (LG) (p < 0.001 for both; Fig. 5(A), (B)). Examining the time series of the TG-DCS BFi vs. the TDF rCBF signal, we observed a greater increase in LG versus EG after the first green line (sedation turned off). At the second green line, a neurological examination was performed that included painful stimulation to test the level of arousal and motor functioning. A sharp increase in CBF was observed in both EG and LG. The magnitude of the change in rCBF triggered a calibration cycle of the TDF device, which is an expected behavior designed to guarantee the accuracy of its measurements (see the data gap beginning ∼ 3500 s). The third green line marks the restart of sedation, which caused a decrease in both TG-DCS BFi and TDF rCBF. It is important to note that these perturbations to the patient can cause systemic changes in blood flow, which may affect both CBF and scalp blood flow simultaneously. To show the feasibility of TG-DCS at high temporal resolution (50 Hz), wider gates closer to the peak of temporal point spread function (TPSF) were used to increase the signal-to-noise ratio (SNR). The opening time/gate width relative to the peak was adjusted to +200 ps/1800 ps. The data were correlated at a rate of 50 Hz, which is sufficient to resolve heart rate up to 300 beats per minute. At higher resolution, we observed greater fluctuations in the BFi (Fig. 5(C)). When a portion of the data is zoomed in, we observed a pattern like that of the pulsatile flow, which is related to the heart rate ( Fig. 5(D)).
To validate that we can obtain pulsatile blood flow, a second-order zero-phase Butterworth filter with a passband of 0.75-3 Hz (45-180 bpm) was used. The signal was then normalized by first subtracting the mean then dividing it by the standard deviation of itself. Next, Welch's method with 64 windows and 50% overlap was used to obtain the power spectral density (PSD) of the TG-DCS BFi (Fig. 6) [70]. We compared the frequency spectra to the frequencies of the heart rate (HR) signal, which was recorded using a bedside data aggregation system to ensure signals were time-locked (Component Neuromonitoring System; Moberg Solutions, Inc; Ambler, PA). The PSD obtained from TG-DCS was overlaid with the PSD from the HR (Fig. 6). A ρ of 0.89 (LG vs HR) and 0.80 (EG vs HR) (p < 0.001 for both) were obtained showing a significant correlation between TG-DCS and HR signals.

Conclusion
To our knowledge, this work is the first report on the clinical use of TG-DCS using SNSPD at 1064 nm in neuroICU settings. We were able to resolve continuous, pulsatile blood flow from a TBI patient. Pulsatile blood flow measurements are attractive because it can provide cerebral autoregulation-related biomarkers. To obtain pulsatile blood flow, one needs to have higher SNR, which can be obtained by software correlator [71], Forti et al. [37] highly parallelized detection [66], multimode detection schemes implemented in interferometric diffusing wave spectroscopy [65] and using longer wavelengths as shown here as well as other groups [46,47,55,56]. Our compact probe was able to measure BFi, which correlated well with the rCBF obtained from an invasive thermal diffusion-based technique. Non-invasive measurement could provide continuous blood flow data during a period in which invasive measures were limited by the need for signal calibration. Additionally, we were able to adjust the photon gating to manipulate the temporal resolution such that pulsatile blood flow could be resolved, which was verified against the heart rate recorded using a commercial clinical device. The ability to adjust the SNR, probing depth, and sampling frequency provided an exceptionally flexible tool that can be customized to the patient and to resolve specific physiological indicators, such as the relationship between arterial pressure and cerebral blood flow that could ultimately be used to estimate intracranial pressure or cerebral autoregulation.