2.1.

Design of a Combined DCS-NIRS System

To calculate impedance using pulsatile hemodynamics, we developed a combined DCS-NIRS system, which operated at 500 Hz sampling rate and measures colocalized blood volume and blood flow changes. The source-detector distance was set to 2 cm. The DCS device is based on a software correlator approach and uses a long coherence length fiber-coupled laser operating at 785 nm (DL785-70-3O, Crystalaser, Reno, Nevada, United States). The output power was reduced to 30 mW using a manual fiber attenuator (VOAMMF, Thorlabs, Newton, New Jersey, United States). The light was coupled to a $400\text{-}\mu \mathrm{m}$ multimode (MM) fiber for the illumination of tissue. Collection of back-scattered light was done using a bundle of four “few-mode” (FM) fibers (980HP, Thorlabs) to increase SNR²⁷ and detected with a four-channel single photon avalanche detector (SPAD) module (SPCM-AQ4C, Excelitas, Waltham, Massachusetts, United States). The digital pulses from each detector are received by an field programmable gate array (FPGA; XC7A100T-2CSG324C, Xilinx, San Jose, California, United States). The FPGA is internally configured as an asynchronous counter that increments an internal 3-bit counter, per channel. The value of the asynchronous counter is recorded at 10 MHz, yielding a bin size of 100 ns.

The NIRS detection was colocalized with the DCS, using the same laser for illumination. A $200\text{-}\mu \mathrm{m}$ MM fiber was placed tangential to the DCS detector fibers where the 2.5-mm ceramic ferrules were touching. The distance between the centers of the DCS fiber bundle and NIRS detector fibers was 2.5 mm and thus probes the same location as DCS for diffuse applications. The light was detected using a temperature-compensated avalanche photodetector (APD410A, Thorlabs). A custom analog front-end (AFE) amplified and low-pass filtered the signal with a cut-off frequency of 100 Hz. Two detectors and two AFEs were used to allow for averaging of multiple detectors together, or for simultaneous measurements at two different detector locations depending on the experiment. These signals were time-interleaved using an analog multiplexer (MUX), then digitized using a 14-bit analog-digital converter (ADC) operating at the same 10-MHz FPGA sampling clock. A high-level schematic of the DCS NIRS device and probe design is shown in Fig. 1. Two identical optical probes were built for simultaneous acquisition of the periphery and the head.

Fig. 1

(a) High-level system block diagram of the combined DCS-NIRS system. A laser source with an attenuator coupled to an MM fiber is shown in blue. NIRS signals are received through an MM fiber, filtered using a custom AFE then time-interleaved using an MUX, and sampled using a single ADC shown in orange. DCS signals are received through an FM fiber bundle and sent to the SPAD array. The SPAD detectors emit an electrical pulse that is then buffered and counted using an FPGA shown in green. Data are placed on a bus to be sent to the computer shown in gray. (b) Probe design schematic showing relative fiber positions and spacing.

DCS and NIRS data were collected and streamed to a computer using a USB 3.0 controller configured as a high-speed acquisition system. The raw DCS and NIRS data were recorded at 10 MHz for postprocessing. The instrument also contained a custom electrocardiogram (ECG) device that recorded the onset of the R-peak of the ECG waveform. The R-peaks are the highest amplitude component of the ECG waveform and signify the onset of ventricular depolarization.²⁸ The timing information provided by the ECG R-peaks allowed for averaging several ($\sim 100$) blood flow and volume pulses to increase the fidelity of the cardiac waveform.

2.2.

Human Subject Experiments

Flow impedance is the frequency-dependent analog of flow resistance and can be calculated as

Although we could measure pulsatile blood flow using our DCS device, we did not have a way to directly measure pulsatile blood pressure changes in the same location. To address this, we first ran a study to investigate the relationship between pulsatile pressure changes and pulsatile blood volume changes measured with NIRS in the arm. We recruited four healthy human volunteers and measured a linear relationship between changes in blood volume and pressure at the cardiac pulsation due to the constant elastic modulus of vessels over small distentions.²⁹

After validating this relationship, we used NIRS-based pulsatile volume changes as a surrogate for pulsatile pressure changes in a second experiment to measure cerebral impedance. We measured impedance at the cardiac frequencies and harmonics on six healthy volunteers.

2.2.1.

Experiment 1: pulsatile blood pressure versus volume in the arm

The experimental protocol was approved by the Carnegie Mellon University Institutional Review Board (IRB #2015 00000113). Written informed consent was obtained from all participants before the study. The DCS-NIRS probe was placed on the forearm above the brachioradialis muscle, as seen in Fig. 2. Only the NIRS system was activated to measure pulsatile blood volume changes. A continuous noninvasive blood pressure monitor (CNAP 500, CNSystems, Graz, Austria) was placed on the finger of the same arm to record pulsatile blood pressure. Our ECG monitor was placed on the chest to provide the timing of the cardiac cycle. A separate blood pressure cuff was placed on the same arm to modulate the pulsatility in blood pressure through partial arm occlusions. The pressure cuff was inflated such that the pulsatility (difference between systolic and diastolic blood pressure) reported by the CNAP blood pressure monitor was approximately 10 mmHg for 2 min, followed by 20 and 30 mmHg. Baseline measurements of 5 min at the subjects’ nominal MAP were taken before and in between cuff inflation.

Fig. 2

Experimental setup and probe placement for validation of pulsatile pressure-volume relationship. A pressure cuff was placed on the arm to modulate pulsatile pressure through arm occlusions. A DCS-NIRS probe and a noninvasive pressure monitor were placed on the arm and finger, respectively to record hemodynamics. ECG leads (not shown) were placed on the chest to provide a timing signal for postprocessing.

2.2.2.

Experiment 2: impedance changes in head and arm

Under the same IRB protocol, an additional six healthy volunteers were enrolled in a HOB tilting study. Gravity and posture changes can perturb CA through baseline changes in CPP, thereby causing a change in impedance in the head.^30–34 Using an inverting table, subjects initially lay flat at 0 deg for 15 min, followed by a full body tilt downwards to $-40\mathrm{deg}$ for 15 min, then returning to 0 deg for another 15 min. A DCS-NIRS probe was placed on the forehead, 2 cm above the left eyebrow, to measure pulsatile blood flow and volume changes. An additional, identical DCS-NIRS probe using a second 785-nm laser was placed on the arm, above the brachioradialis muscle. The arm location served as a control measurement because CA does not exist in the periphery. Thus no impedance changes were expected in the arm. The experimental setup is shown in Fig. 3.

Fig. 3

Experimental setup for impedance measurements. Subjects’ entire bodies were tilted head down to induce impedance changes in the head. A DCS-NIRS probe was placed on the forehead, 2 cm above the left eyebrow, and the arm above the brachioradialis.

2.3.

Data Analysis

After the DCS-NIRS data were streamed to a computer, data were processed using a standard Linux-based computer running the Python programming language (python.org). The custom Python program saves the unprocessed binary data from the device directly to the disk while performing temporal autocorrelations and curve fitting and displaying the resulting DCS flow and NIRS data at $>50\mathrm{Hz}$.

2.3.3.

CVI calculation during HOB tilting (experiment 2)

The data processing approach for impedance extraction is shown in Fig. 4. We restricted our analysis only to the cardiac frequencies such that we could assume the subjects’ autoregulation system is stable. We applied an eighth-order Butterworth filter with a cutoff frequency of 0.35 Hz to remove the influence of respiration and spontaneous fluctuations in both BFI and HbT measurements. The last 8 min of data per tilting angle were used for further processing to reduce the initial transient effects of posture changes. Data were segmented into 30-s windows with a 29-s overlap. Based on the results of experiment 1, we used pulsatile HbT as a surrogate measurement for pulsatile pressure changes. The Fourier transform of each time window was applied to obtain the spectral power of pressure (using HbT) and BFI. Impedance was calculated by applying Eq. (1) to each time window.

Fig. 4

Overview of signal processing from DCS-NIRS data to impedance per angle. DCS and NIRS signals were high-pass filtered to remove unwanted signals and then separated into overlapping 30-s windows. The Fourier transform was taken for each window, and all windows were averaged together to determine the spectral flow and pressure, respectively. Impedance was calculated by the ratio of pressure and flow.

We calculated the impedance spectra at the cardiac pulsation frequency and its harmonics, and we linearly interpolated the spectral data over the frequencies where there is no cardiac spectral energy. To determine the spectral regions to interpolate, we defined an SNR metric

Eq. (5)

$${\mathrm{SNR}}_Z(\omega )=\frac{\overline{|Z(j\omega )|}}{\sigma (|Z(j\omega )|)},$$

where $\overline{|Z(j\omega )|}$ is the average magnitude of impedance, averaged over all time windows, and $\sigma (|Z(j\omega )|)$ is the standard deviation of the magnitude over all time windows. We replaced frequencies of ${\mathrm{SNR}}_{\mathrm{Z}}(\omega )<1$ with interpolated spectral content between cardiac harmonics.

Because we did not have access to absolute CPP, we calculated relative changes in impedance, rather than absolute impedance curves. To allow for inter-subject comparisons, subject-dependent scaling factors, such as the pulsatile pressure-volume coefficient, must be normalized out from each subject’s set of impedance curves. This was done by dividing all the curves from a particular subject by the subject’s impedance at the heart rate frequency for the initial 0 deg angle. The location of this fundamental heart rate frequency was determined by taking the location of the average peak of the blood flow spectra across the 30 s windows at the 0 deg angle.

Impedance was also visualized in the time domain as a conductance impulse response defined as

Eq. (6)

$$q(t)=p(t)*g(t),$$

where $q(t)$, $p(t)$, and $g(t)$ are temporal flow, pressure, and the conductance convolutional kernel (impulse response), respectively. This convolutional kernel can be derived from impedance as

Eq. (7)

$$g(t)=F^{-1}\left\{\frac{1}{Z(j\omega )}\right\}.$$

Taking the average impedance spectra without interpolation, we applied Eq. (7) to estimate $g(t)$. We then applied an eighth-order Butterworth low-pass filter with a cutoff frequency of 35 Hz to remove high-frequency noise artifacts caused by our signal processing from our estimate. We used the spectra found prior to interpolation to avoid discontinuities in the spectra which would cause artifacts when taking the inverse Fourier transform.

3.1.

Relationship Between Pulsatile Blood Pressure and Volume Changes (Experiment 1)

An example of averaged pulsatile blood flow and blood volume data is seen in Fig. 5 showing DCS and NIRS measurements in the head and arm averaged over approximately 100 cardiac cycles. Data were aligned to the ECG R-peak and averaged. The high sampling rate of our DCS-NIRS device provides sufficient temporal resolution to resolve the dicrotic notch in the ECG-averaged cerebral waveforms, as seen in Fig 5.

Fig. 5

Sample DCS-NIRS curves for head and arm. Data show the resulting average of $\sim 100$ individual cardiac pulsations sampled at 500 Hz. The shaded area represents the standard deviation from the mean.

Because impedance is defined as the spectral ratio between blood pressure across the vessel and blood flow, we estimated pressure changes by assuming a linear relationship to blood volume changes. To show this, we quantified the magnitude of the pulsatile blood pressure and the magnitude of pulsatile blood volume in the periphery. The relationship between the two is seen in Fig. 6, which shows a scatter plot of the magnitude of blood volume changes as a function of blood pressure changes. The results for three subjects are shown; data from the fourth subject was rejected based on our SNR criteria. At the baseline blood pressure, all subjects showed a pulsatility of 30 to 40 mmHg. A linear relationship was fitted between pulsatile blood pressure and blood volume for the three subjects resulting in $R^2$ values of 0.79, 0.96, and 0.88 for each subject. Because our system cannot resolve absolute hemoglobin values, changes in pulse magnitude are relative and slopes between subjects are subject-specific. Based on the linearity found in Fig. 6, we used $\mathrm{\Delta }\mathrm{HbT}(\mathrm{t})$ as a surrogate for $\mathrm{\Delta }\mathrm{P}(\mathrm{t})$ for further analysis.

Fig. 6

Scatterplot of pulsatile volume changes as a function of pulsatile pressure changes. A linear fit was performed for each subject and $R^2$ was determined. Error bars denote the standard deviation from the mean.

3.2.

Cerebrovascular Impedance and Impulse Response During HBO Tilting (Experiment 2)

Using the linear relationship between blood pressure and blood volume at the cardiac pulsation found in experiment 1, we calculated CVI for each tilt angle based on Eq. (1), using $Q=\mathrm{BFI}$ and $P=\mathrm{\Delta }\mathrm{HbT}$. Figure 7(a) shows the group average impedance curves across all subjects ($N=6$), with error bars representing the standard error from the mean. There was a negligible change in the heart rate during the experiment, with the average heart rate across all six subjects at the baseline being 61.35 BPM, which increased to 61.49 BPM during the $-40\mathrm{deg}$ tilt and decreased to 61.44 BPM during the recovery. Average cerebral BFI increased by 140% at $-40\mathrm{deg}$ relative to the initial baseline, then decreased to 108% when returning to 0 deg. The results show that the average impedance across subjects between 1 and 2 Hz is lower in the head at $-40\mathrm{deg}$ when compared to the 0 deg angles. In contrast, the arm impedance spectra show minor changes over HOB tilting angles [Fig. 7(a)].

Fig. 7

(a) Normalized group averaged ($N=6$) impedance spectra and (b) conductance impulse response for the head and arm DCS-NIRS locations. Each color represents a different HOB angle. Shaded areas signify standard error from the mean.

Figure 7(b) shows the normalized group averaged conductance impulse response for each detector location. Like the trends in the impedance spectra, the data show an increase in impulse amplitude in the response curve of the head when tilted downwards to $-40\mathrm{deg}$ when compared with the 0 deg angles. Likewise, there is little morphological change in the impulse response in the arm when tilting.

4.1.

Design of a High-Speed DCS-NIRS System

Our combined high-speed DCS-NIRS system can simultaneously measure pulsatile, colocalized blood volume, and blood flow. Using ECG-timed pulse averaging, we could further increase the fidelity of hemodynamic pulsatile signals. Such high-resolution signals can subsequently allow for deeper insight into tissue’s physiological status. Several prior studies have investigated the combination of DCS and NIRS to measure cerebral hemodynamics.^48–53 However, many of these studies employed the use of hardware correlation and/or frequency domain NIRS, which restrict sample rates to $\sim 1$ to 10 Hz, which may not provide sufficient temporal bandwidth to reconstruct the higher-order harmonics of the cardiac signals and thus cannot resolve the finer details of the cardiac waveforms. Our device, in combination with ECG alignment, can provide sufficient temporal resolution and SNR to analyze several harmonics of the pulsatile waveform. Furthermore, many DCS-NIRS implementations rely on separate DCS and NIRS devices, which present alignment and optical crosstalk challenges which limit temporal resolution.⁴⁸ By combining DCS-NIRS hardware into a unified package and sharing a common laser source, our device circumvents alignment and optical crosstalk challenges, which allow reliable reconstruction of pulsatile hemodynamics.

4.2.

Local Blood Pressure Estimation

For this work, we have assumed a linear relationship between pulsatile blood volume and pulsatile blood pressure changes based on data collected in the arm, with data shown in Fig. 6. This indicates a constant elastic modulus for the blood vessel bed during a cardiac pulsation.²⁹ Pulsations in the NIRS signal are due to volume expansions in the vessels from pulsatile pressure changes. In the arm, these pulsatile volume changes are due to pulsations in ABP. However, in the head, the effects of ICP must be considered. Pulsatility in ICP is a passive response to pulsations in ABP.^54,55 These pulsations in ICP can reduce the magnitude of the CPP pulsations, which drive vessel dilation in the brain. Due to the passive nature of pulsations in ICP, we assumed that NIRS-reported pulsatile blood volume change is linear to CPP. Therefore, we assumed that the linearity between blood volume and CPP pulsations still holds for cerebral measurements. Although the pulsatile pressure-volume slope in the brain may differ from the arm, the normalization method in CVI calculation removes the subject-dependent slope and is therefore still valid.

Exponential relationships in vessels can occur at extreme pressure and volume changes where vessel elastic-modulus is no longer constant.²⁹ To ensure that our impedance trends were still consistent using a nonlinear pulsatile pressure-volume relationship, we ran our analysis assuming a power law relationship ($\mathrm{\Delta }P=k^{\mathrm{\Delta }\mathrm{HbT}}$ where $k=10$). We found (data not shown) that the impedance curve trends with tilting angle were qualitatively the same as when making the linearity assumption. This shows that our method is not confounded by a nonlinear pressure-volume assumption.

4.3.

Agreement of Impedance and Autoregulation in Posture Changes

Due to the high sampling frequency for our NIRS and DCS device, harmonics of the pulsatile hemodynamics could be used for the estimation of vascular impedance. Posture changes have been shown to affect CPP, which trigger an autoregulative response^3,56 and subsequently induce impedance changes. Literature shows that CPP decreases in Trendelenburg positioning (head down tilt).^32,33,57 At a $-40\mathrm{deg}$ tilt, we expected a decrease in CPP, which would cause vasodilation in the cerebral microvasculature, thereby causing a decrease in CVI when compared to 0 deg. In contrast, we expect that HOB tilting would not affect the CVI in the periphery. Our results shown in Fig. 7(a) resemble previously reported trends in larger vessels where vasoconstriction leads to an overall increase in impedance and, conversely, vasodilation results in a lowering of impedance.^3,13 Thus, it corroborates the fact that we observe a decrease in impedance when tilted downwards as vessels must dilate to maintain sufficient perfusion in the face of decreasing CPP.

4.4.

Limitations and Future Work

We have shown that changes in impedance can be recorded as a function of the HOB tilting angle. However, this method is unable to quantify absolute impedance due to the subject-specific pressure-volume proportionality coefficient, which limits the estimation of absolute pulsatile pressures. This may be caused, in part, using continuous-wave NIRS, which is only able to measure blood volume changes rather than absolute values. Despite our impedance estimation being a relative measurement, we have found that normalizing the impedance curve using the heart rate frequency allows for inter-subject comparisons.

For this work, we assumed our HOB tilts induced changes in CPP, which triggered a CA response. However, our study is not able to directly quantify these CPP changes without the use of invasive pressure probes. Future controlled invasive experiments using ICP and ABP pressure probes are needed to measure CPP and quantify the relationship between CPP and CVI. Furthermore, our approach is only applicable when we can assume that impedance is quasi-constant throughout each tilt angle measurement. This notably has implications for the impedance frequencies we can estimate. In the brain, low-frequency changes are modulated by CA.^9–12 Thus, the impedance at these frequencies will vary over time and cannot be estimated using our method. However, our approach assumes that the cardiac frequencies are not actively regulated through active vasodilation or constriction to counteract pressure changes. We therefore can only estimate vascular impedance at higher frequencies such as the cardiac pulsation.

Finally, our device is operating at a single wavelength close to the isosbestic point, namely at 785 nm.⁴⁴ Whereas this operation allows for an estimation of total hemoglobin changes, we were not able to differentiate between oxyhemoglobin and deoxyhemoglobin. A future iteration of this device will incorporate multiple wavelengths to extract information about different hemoglobin species. To achieve this, a multicore source fiber could be used for illumination where each core contains a separate wavelength. At the detector end, the DCS fibers and two NIRS fibers could be bundled together to enable colocalization of the NIRS and DCS signals across the different wavelengths and optical filters used for wavelength selectivity.