2.1.

Instrumentation

Kernel Flow^11,12 is a trNIRS device that provides high-density, full-head coverage. The system is equipped with 52 hexagonal modules (Fig. 1), each having 2 laser sources in the middle, emitting at 690 and 850 nm, and 6 detectors at the vertices located 10 mm from the sources. There are 52 source locations and 312 detectors. The system provides up to 2206 potential channels with varying source–detector distances (SDDs); 312 are short-distance channels (SDD = 10 mm), and 1894 are long-distance channels (SDD between 20 and 60 mm) across the head. For this study, intramodular channels were classified as short channels (SDD = 10 mm), and cross-modular channels with $\mathrm{SDD}\ge 25\mathrm{mm}$ were classified as long channels. The placement of the optical probes is divided into eight plates that roughly cover the brain’s four lobes and will be referred to as frontal, temporal, occipital, and sensorimotor [Fig. 1(d)]. Additionally, the system uses spring-loaded light pipes for transmission and detection to improve scalp coupling. The device weighs 2.2 kg and has a sampling rate of 7.1 Hz with a built-in temperature sensor and accelerometer.

Fig. 1

(a) Kernel Flow system showing the inside of the helmet. (b) One of the Kernel Flow modules with the light source in the middle and detectors around the perimeter. (c) Kernel Flow on a participant’s head. (d) Kernel Flow probe placement on a standard head model.²³ Red dots depict laser sources, and blue dots are detectors. The black circles show the division of the eight brain regions (two hemispheres, each divided into frontal, sensorimotor, temporal, and occipital region). Images in panels (a) and (b) are adapted from Ref. 24.

2.2.

Experimental Protocols

2.2.2.

Protocol 2: Phenylephrine administration during shoulder surgery

Adult patients undergoing elective shoulder surgery at St. Joseph’s Hospital (London, Ontario) were prospectively recruited between October 2022 and March 2023. All procedures were approved by the Health Sciences Research Ethics Board at Western University (REB# 115021), which adheres to the guidelines of the Tri-Council Policy Statement for research involving humans. Participants (Table 2) provided written informed consent following written and verbal explanations of the protocol. Once the patient was anesthetized, intubated, and placed in the beach chair position, the Kernel Flow was placed on the head and stabilized with a head strap (Fig. 2). To avoid overheating, a forced-air cooling system (Bair Hugger, 3M Medical) was used to blow cool air at the Flow device. Data were collected continuously throughout the surgical procedure, and instances of PE administration were noted. MAP and heart rate (HR) measurements were acquired using hospital equipment every 5 min. To investigate the effect of PE, 1 min prior to the PE administration was used as the baseline, and 5 min post-injection were included in the analysis.

Table 2

Demographics of the patients included in the analysis.

Patient ID	Surgery side	Sex	Age	Weight (kg)	Height (cm)	Shoulder surgery type	Number of PE bolus injections
5	R	F	75	69	160	Arthroplasty reverse	6
7	L	M	69	92	173	Arthroplasty reverse	4
12	R	M	66	92.5	175	Shoulder arthroplasty total	2
13	R	M	68	115	180	Arthroplasty reverse	3
23	R	F	60	67.5	147	Arthroplasty reverse	1

Fig. 2

A shoulder surgery patient with the Kernel flow system on the head. The device was stabilized with a head strap.

2.3.

Data Processing

Raw data were uploaded to Kernel’s cloud-based pre-processing pipeline, which starts by removing poor-contact channels based on the shape of the distribution of times-of-flight (DTOF), photon counts, scalp coupling index, and HR peak power.¹² After preprocessing, we received the pruned data containing only “good” quality channels. The pipeline then detects the peak of each DTOF and defines an integration window for computing the statistical moments: 10% of the peak on the rising edge and 1% on the trailing edge of the histogram. Thereafter, three statistical moments are calculated: (1) the total number of photons ($N$, i.e., the zeroth moment); (2) the mean time of flight ($⟨t⟩$, i.e., the first moment); and (3) the variance ($V$, i.e., the second moment). Note that higher moments are more sensitive to the brain, whereas lower moments are more sensitive to the scalp.^15,26 For the remainder of the paper, the total number of photons ($N$) at a short SDD will be referred to as $N_{\text{short}}$, and the total number of photons at a long SDD labeled $N_{\text{long}}$. The same nomenclature will be applied to the mean time of flight ($⟨t⟩$) and the variance ($V$). The least depth-sensitive measurement is $N_{\text{short}}$, which predominantly reflects the scalp response. Likewise, measurements with the greatest depth sensitivity are derived from $V_{\text{long}}$, which mostly reflects the brain response.^16,27

Data were further processed using an in-house pipeline developed in MATLAB (Mathworks Inc., United States). Changes in each moment relative to its baseline value were calculated to generate three different time courses for each wavelength. Time courses were then converted into corresponding absorption changes ($\mathrm{\Delta }{\mu }_a(\lambda )$) using sensitivity analysis^28,29 [Eq. (1)], where $\mathrm{\Delta }S$ is the signal change for a given moment and ${\mathrm{SF}}_S$ represents the corresponding sensitivity factor (which also dependents on the source-detector distance)

$\mathrm{\Delta }{\mu }_a$ was then converted to oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) concentration changes ($\mathrm{\Delta }[\mathrm{HbO}]$ for HbO, and $\mathrm{\Delta }[\mathrm{Hb}]$ for Hb) using Eq. (2), where ${\epsilon }_{\mathrm{HbO}}(\lambda )$ and ${\epsilon }_{\mathrm{HbO}}(\lambda )$ are the molar extinction coefficients for HbO and Hb, respectively

Baseline values of [HbO] and [Hb] (i.e., ${[\mathrm{HbO}]}_{\mathrm{b}}$ and ${[\mathrm{Hb}]}_{\mathrm{b}}$) from the literature were assumed for all participants (33 and $24\mu \mathrm{M}$,^30,31 respectively). Relative concentration changes, calculated from the different moments and SDDs, were used to estimate ${\mathrm{StO}}_2$ changes using the following equation:

Detection channels were sorted into eight brain regions based on the four brain lobes of each hemisphere, and regional averages were calculated. Time courses were filtered with a zero-phase filter using a window size of approximately 1.5 s.

2.4.

Correction for Temporal Drift

The time course of $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ exhibited a strong dependence on the device’s temperature, which manifested as temperature-dependent signal drifts. This was particularly evident during the longer monitoring periods in the surgery study (protocol 2). To mitigate the confounding effects of the temperature drift, dynamic bandpass wavelet filtering was implemented. The filtering consisted of applying a continuous wavelet transform to decompose the temperature time course of each module to identify the frequency of temperature drift with maximum power across the monitoring period. This frequency was then used as the lower limit of the bandpass filter for the $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ time courses. Across all patients, the lowest frequency of the bandpass filter was 0.008 Hz (period of 125 s). The upper limit of the bandpass filter was consistently 3.1 Hz, as limited by the device sampling rate (7.1 Hz). Each channel was filtered with its module-specific, time-varying bandpass filter. This resulted in 52 unique filters used across the head. This approach could not be applied to the DCC time courses (protocol 1) because the measurements were only 90 s; instead, a simple first-order polynomial was used for detrending.

2.5.

Statistical Analysis

All data were assessed for normality through a visual inspection of a QQ plot as well as the Shapiro–Wilk test. All statistics were conducted in GraphPad Prism version 9 (GraphPad Software Inc., San Diego), and statistical significance was defined as $p\le 0.05$.

3.1.

Protocol 1: Digital Carotid Compression

Ten subjects were included in the study ($28\pm 5$ years; five females and five males; Table 1). For the frontal region, $36\%\pm 23\%$ of the long-distance channels had good-quality signal, among which $52\%\pm 16\%$ were between 25 and 35 mm, $39\%\pm 8\%$ between 35 and 45 mm, and $8\%\pm 12\%$ between 45 and 60 mm. However, the number or good-quality channels decreased significantly ($p\le 0.004$) for the other regions: $8\%\pm 12\%$ for the sensorimotor cortex, $11\%\pm 17\%$ for the temporal lobe, and $5\%\pm 10\%$ for the occipital lobe [Fig. 3(a)]. Additionally, no significant interaction was found between hair type and region ($p=0.3$). However, significant main effects of hair type ($p\le 0.0001$) and region ($p\le 0.0001$) were found. Subjects with “light” hair ($n=4$; three males, one female) had a significantly higher percentage of good-quality channels [$27\%\pm 25\%$ versus $7\%\pm 10\%$; $p\le 0.001$; Fig. 3(b)] when compared to those with “dark” hair ($n=6$; two males, four females).

Fig. 3

(a) Regional sensitivity of the device ($n=10$; mean age: $28\pm 5$; five females). Box-and-whisker plots show first and third quartiles as outlined by the box, with the middle line indicating the median and the ends of the line indicating minimum and maximum values. Crosses indicate outliers. Asterisks indicate that the frontal region was significantly different from all other regions. (b) “Light” ($n=4$) and “dark” ($n=6$) hair participants’ regional sensitivity are displayed as stacked bars of the four regions. An example of the active channel maps overlaid on a standard head model²³ from a participant from the “light” hair group is shown in panel (c), and an example of the “dark” hair group is shown in panel (d).

Given that the frontal was the only region that had good-quality long-distance channels for all the participants [see Figs. 3(c) and 3(d)], it is the only region that could reliably be used to assess differences in the response to DCC across all participants. Figure 4(a) shows $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ responses to DCC obtained from the zero-order moment measured at the shortest SDD, $N_{\text{short}}$ (i.e., the measurement that is overwhelmingly sensitive to the extra-cerebral tissues). Figures 4(b)–4(d) display the hemodynamic responses for all three moments extracted from data measured at $\mathrm{SDD}\ge 25\mathrm{mm}$. The panels are ordered from the least to the most depth-sensitive measurements [moving from Figs. 4(b)–4(d)]. The contralateral time courses are in Fig. S1 in the Supplementary Material but show negligible changes during DCC. Further analysis of the ipsilateral $\mathrm{\Delta }[\mathrm{HbO}]$, $\mathrm{\Delta }[\mathrm{Hb}]$, and $\mathrm{\Delta }{\mathrm{StO}}_2$ profiles across time revealed a significant interaction in all three parameters across the moments ($p\le 0.0001$ for $\mathrm{\Delta }[\mathrm{HbO}]$, $\mathrm{\Delta }[\mathrm{Hb}]$, and ${\mathrm{StO}}_2$)—see Fig. S2 in the Supplementary Material for the complete analysis of significant differences. The most notable differences were the faster recovery during desaturation in the second half of the DCC and the hyperemic response after DCC for $V_{\text{long}}$.

Fig. 4

(a)–(d) Average $\mathrm{\Delta }[\mathrm{HbO}]$ (red) and $\mathrm{\Delta }[\mathrm{Hb}]$ (blue) responses to DCC (indicated by the gray-shaded region) in the ipsilateral frontal region. Time courses are presented for the signals measured at short (a) and long (b)–(d) distances. Responses derived from (a) and (b) the total number of photons ($N$), (c) mean time of flight ($⟨t⟩$), and (d) variance ($V$) are shown. (e)–(g) Average long distance (black) and short distance (green) $\mathrm{\Delta }{\mathrm{StO}}_2$ derived from (e) $N$, (f) $⟨t⟩$, and (g) $V$ are shown. All time courses were averaged across 10 subjects and shading surrounding each line represents the standard deviation.

3.2.

Protocol 2: Phenylephrine Administration During Shoulder Surgery

Data were acquired from 27 shoulder surgery patients. All patients were classified as light-skinned (I or II on the Fitzpatrick skin type scale) and had thinner lighter-colored hair due to age and demographics. Wearing the Kernel flow helmet resulted in a minor skin abrasion for one patient, but no complications were reported for the other subjects. The average age was $65\pm 15$ years (14 females and 13 males). Seventeen patients had right shoulder surgery, and ten underwent left shoulder surgery. Only 10 patients received at least one bolus injection of PE, and of these, 5 were excluded due to technical challenges, such as motion artifacts or corrupt files. For consistency, only the first PE injection from each patient was analyzed and compared. Table 2 provides the demographics of the five patients ($68\pm 5$; two females, two males) included in the analysis and the number of PE injections they received.

To assess the device’s regional sensitivity, all 27 patients were considered. Four patients were excluded from this analysis due to technical issues where some modules could not be connected. For protocol 2, the signal was generally high as the subjects were older and had lower hair density. Overall, the frontal region had $67\%\pm 37\%$ of the long-distance channels classified as good quality, the sensorimotor had $71\%\pm 32\%$, the temporal had $59\%\pm 35\%$, and the occipital had $62\%\pm 32\%$. No significant differences were found ($p=0.5$) in the percentage of good-quality channels between brain regions in this patient population. Resultantly, limiting long distance channels to those with SDD of 25 to 35 mm was suitable for the rest of the analyses.

An example of temperature-drift correction for each of the three moments is shown in Figs. 5(a)–5(c). In this case, initial $N_{\text{long}}$ and $V_{\text{long}}$ time courses followed the same trend as temperature, whereas ${⟨t⟩}_{\text{long}}$ showed the opposite trend. Figures 5(d) and 5(e) present the average $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ responses to phenylephrine injection before and after temperature-drift correction. The results demonstrated that temperature correction reduced the variability across subjects, as indicated by the standard deviation, and, more importantly, resulted in the $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ time courses returning to baseline.

Fig. 5

Example time courses of (a) $N_{\text{long}}$, (b) ${⟨t⟩}_{\text{long}}$, and (c) $V_{\text{long}}$ from one module over the frontal region of a patient. The overlaid colored boxes indicate windows with different wavelet filter parameters that varied with the corresponding temperature time course (dashed black). For each moment, both the raw and filtered time courses are shown for [HbO] (magenta and red, respectively) and [Hb] (cyan and blue, respectively). Average [HbO] and [Hb] time courses in response to a phenylephrine injection are shown (d) before (e) and after temperature correction. These time courses were calculated from $V_{\text{long}}$. The baseline is displayed in gray, and the response to the PE injection is in white.

Figure 6 shows the average time courses for the frontal region for patients included in the PE analysis ($n=5$). The PE injection occurred at the 1-min time point. Through qualitative observation, Fig. 6(a), predominantly the scalp response, showed large variability in the data; however, there appears to be a sharp decrease in $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ following PE injection. Qualitatively, in the more brain-sensitive measurements [$V_{\text{long}}$, Fig. 6(d)], there is a pronounced desaturation with a simultaneous decrease in $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ increase. Further, the analysis of the time courses of Δ[HbO], $\mathrm{\Delta }[\mathrm{Hb}]$, and $\mathrm{\Delta }{\mathrm{StO}}_2$ (see Fig. S3 in the Supplementary Material) revealed a significant main effect of time in $\mathrm{\Delta }[\mathrm{HbO}]$ only ($p\le 0.04$) as the baseline was significantly different from the injection period.

Fig. 6

(a)–(d) Average $\mathrm{\Delta }[\mathrm{HbO}]$ (red) and $\mathrm{\Delta }[\mathrm{Hb}]$ (blue) responses to the first phenylephrine bolus injection (indicated by the white region; the baseline is shown in gray) in the left frontal region. Time courses are presented for the signals measured at short (a) and long (b)–(d) Responses derived from (a) and (b) the total number of photons ($N$), (c) mean time of flight ($⟨t⟩$), and (d) variance ($V$) are shown. Average long distance (black) and short distance (green) $\mathrm{\Delta }{\mathrm{StO}}_2$ derived from (e) $N$, (f) $⟨t⟩$, and (g) $V$ are shown. All time courses were averaged across five subjects and shading surrounding each line represents the standard deviation.

Moreover, MAP increased post-injection ($61\pm 10\mathrm{mmHg}$ to $80\pm 12\mathrm{mmHg}$; $p\le 0.01$) versus pre-injection, but HR did not significantly change after PE injection ($58\pm 10\mathrm{bpm}$ to $54\pm 11\mathrm{bpm}$; $p=0.08$).

Figure 7 displays the average time courses ($n=5$) in all brain regions, calculated using $V_{\text{long}}$. The curves show that all regions had a similar response to the PE injection, characterized by a simultaneous decrease in $\mathrm{\Delta }[\mathrm{HbO}]$ and increase in $\mathrm{\Delta }[\mathrm{Hb}]$. There were no significant differences in the regional responses to PE injection for the magnitude of $\mathrm{\Delta }[\mathrm{HbO}]$ ($p=0.7$) or $\mathrm{\Delta }[\mathrm{Hb}]$ ($p=0.7$) or time of maximum change ($p=0.99$).

Fig. 7

Concentration changes (ΔC) of HbO (red), and Hb (blue), calculated using $V_{\text{long}}$, showing the average ($n=5$) first response to phenylephrine bolus injection during surgery across the four brain regions (frontal, sensorimotor, temporal, and occipital). The shading indicates the standard deviation across participants.

4.1.

Protocol 1: Digital Carotid Compression

In the DCC protocol, the device’s sensitivity was assessed in young healthy subjects by calculating the percentage of good-quality channels across the head. The analysis revealed that the frontal region had a significantly higher number of good-quality channels compared to the sensorimotor, temporal, and occipital regions. Additionally, the device had better sensitivity in subjects with light hair than those with dark hair. These findings indicate that signal quality depends on the type and density of hair. Consequently, using the device for full-head neuromonitoring in patients with darker and longer hair would be challenging. Further, the regional distribution of channel sensitivity reported in this work showed that the frontal region had the most signal while the occipital region had the lowest percentage of good-quality channels. This is consistent with the findings of Ortega-Martinez et al.,³² who reported a similar sensitivity distribution with the Kernel Flow. Due to the lower overall signals in the younger participants recruited for protocol 1, a large range of SDD values was used in the analysis. The similarity in the time courses for SDDs 25 to 35 mm and those 35 to 45 mm (see Fig. S4 in the Supplementary Material) indicates that this approach was reasonable. Ideally, improving the signal-to-noise ratio would enable a narrower range of SDD to be selected.

Since the frontal region (measured through the forehead) consistently had the best signal in all participants, this region was used to assess the device’s depth sensitivity in the DCC protocol. Additionally, only ipsilateral hemisphere time courses were analyzed, as the contralateral measurements showed negligible changes during DCC, as expected. The hemodynamic responses to the DCC challenge computed using the various depth-sensitive measurements differed significantly. The measurement known to have the most sensitivity to the brain ($V_{\text{long}}$) showed a faster recovery from desaturation in the second half of the DCC, which may be attributed to collateral flow in the brain that is not possible in the scalp. The larger decrease in saturation observed in $N_{\text{long}}$ compared to $N_{\text{short}}$ agrees with our previous work¹⁸ using a cwNIRS system, which found a larger reduction in saturation in the long SDDs compared to the short SDDs in response to DCC. The long-distance cwNIRS measurements were previously attributed to the brain; however, the smaller decrease in saturation in $V_{\text{long}}$ observed in this study suggests that cerebral desaturation following PE administration is likely smaller than previously suggested.¹⁸ Additionally, $\mathrm{\Delta }[\mathrm{Hb}]$ increase was only visible in the long-distance measurements, which indicates greater oxygen extraction in the brain during DCC compared to the scalp, reflecting the lack of oxygen stores and higher energy metabolism in the brain.

Finally, a cerebral ($V_{\text{long}}$) hyperemic response was observed during the recovery in the $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }{\mathrm{StO}}_2$ as blood flow increased following occlusion release. Similar hyperemia has been reported in our previous work using diffuse correlation spectroscopy¹⁸ but was not evident in the ${\mathrm{StO}}_2$ values using cwNIRS. This may be due to the additional depth-sensitivity of $V_{\text{long}}$, as the hyperemia is also not evident in the $N_{\text{long}}$ in the current study. Diffuse correlation spectroscopy provides inherently better depth-sensitivity than cwNIRS due to the substantially higher blood flow in the brain compared to the scalp,^33,34 which can explain this difference. Additionally, these findings agree with Sorteberg et al. who performed DCC on patients with neck neoplasms or traumatic carotid-cavernous fistulas,¹⁷ and found that the blood velocity in the middle cerebral artery (measured using transcranial Doppler ultrasound) showed a hyperemic response following DCC.

4.2.

Protocol 2: Phenylephrine Administration During Shoulder Surgery

There were no significant differences in device sensitivity between brain regions in the surgery data. This finding contrasts with the first protocol in which the frontal region had the best signal. This difference is likely due to the difference in demographics between the two studies. For the surgery study, subjects were older, had lighter-colored hair, and reduced hair density. Therefore, these results agree with the findings of the first protocol that those with “light” hair had better overall signal. The better sensitivity of the device in surgery patients is promising since the aim was to assess the device intraoperatively.³⁵

An unexpected challenge with neuromonitoring over longer periods with the Kernel flow was overheating. An air-cooling device was used to blow cool air on the device; however, there were still temperature-induced signal drifts (see Fig. 5). A dynamic wavelet bandpass filtering approach to remove very low frequencies (below 0.008 Hz or 125 s) was implemented to correct for these signal drifts and very high frequencies (above 3.1 Hz) for motion artifacts. This wavelet filtering proved crucial, as evident by comparing $N_{\text{long}}$, ${⟨t⟩}_{\text{long}}$, and $V_{\text{long}}$ time series before and after applying temperature correction (Fig. 5). It is possible that wavelet filtering may have removed physiological signals outside the bandpass filter. However, this study focused on the rapid cerebral hemodynamic response to a vasopressor, which occurred over a much shorter time frame than those caused by temperature instability.

After 5 min of PE injection, the MAP settles into a new elevated baseline; therefore, it is possible that the hemodynamic parameters settle into a new baseline as well. In this study, however, the filtered $\mathrm{\Delta }[\mathrm{HbO}]$ and $\mathrm{\Delta }[\mathrm{Hb}]$ returned to the pre-PE values, and not a new baseline [see Fig. 5(e)]. This finding agrees with the work of Meng et al.³⁶ and Fassaert et al.³⁷ who likewise demonstrated that while cerebral oxygenation returned to baseline within 5 min, the recovery of basal MAP typically took longer, likely reflecting differences in cerebral and systemic vascular regulation.

The injection of PE caused the expected systemic response, i.e., an increase in MAP consistent with other studies.^38,39 The slight decrease in HR (although not significant) was also expected and has been reported in the literature.³⁹ In terms of the NIRS measurements, $N_{\text{short}}$, which is assumed to predominantly reflect scalp hemodynamics, showed a sharp decrease in $\mathrm{\Delta }[\mathrm{HbO}]$, potentially driven by the vasoconstrictive effects of PE on alpha receptors present in scalp tissue. However, a larger sample size would help investigate this further as the time courses had large variability. Desaturation events, characterized by a simultaneous increase in $\mathrm{\Delta }[\mathrm{Hb}]$ and decrease in $\mathrm{\Delta }[\mathrm{HbO}]$, were observed in the $V_{\text{long}}$, the most brain-sensitive measurement. The cerebral desaturation agrees with previous studies using cwNIRS devices, which report decreases as large as 14% in cerebral oxygen saturation⁴⁰; however, the desaturation reported in the current work are smaller (less than 2%) and likely not clinically significant. Previous studies have shown that the magnitude of ${\mathrm{StO}}_2$ changes in response to PE decreases with increased brain sensitivity.^20,40,41 Additionally, Rosenthal et al.⁴² detected very small changes in cerebral blood flow in response to phenylephrine administration, which is consistent with the small changes in the time-resolved measurements in this work. Our findings also agree with other studies using more depth-sensitive NIRS techniques, such as frequency-domain NIRS (fdNIRS).²¹ It has been suggested that the previously reported cerebral desaturation following PE injection may be related to signal contamination due to extracranial vasoconstriction.⁴³ Importantly, recent work utilizing direct measurements of cerebral tissue oxygen tension in open rat brains⁴⁴ suggests that PE injection does not result in cerebral desaturation. Therefore, despite the well-known superior depth sensitivity of trNIRS,¹⁵ the small magnitude of desaturation observed in $V_{\text{long}}$ may be due to residual scalp contamination. Moreover, the small decreases in $\mathrm{\Delta }[\mathrm{HbO}]$ and increases in $\mathrm{\Delta }[\mathrm{Hb}]$ may be due to changes in cerebral artery vasomotor tone. That is, a small drop in arterial blood volume due to vasoconstriction could shift the measurements to the venous side, leading to the observed trends.

There were no statistically significant differences in the regional response to PE. It is important to note that although there were no significant differences, this may be due to the smaller sample size (only five patients received PE). A subsequent larger study could provide more conclusive answers to potential regional differences in response to PE administration.

4.3.

Limitations

There were a few limitations to this work. First, access to the data in the rawest form (before the automatic processing through the Kernel Flow pipeline) would be helpful to assess why channels were discarded within the automated pipeline and allow us to evaluate the stability and quality of the recorded DTOFs. Additionally, it remains unclear if potential regional differences in hemodynamic responses result from actual physiological differences. They could result from sensitivity differences across the head due to the data processing that uses sensitivity factors to estimate the hemodynamic response. Using magnetic resonance imaging, Wu et al. found that the skull thickness varies across the head;⁴⁵ however, this was not accounted for in the sensitivity factors and could impact our conclusion on the lack of regional differences.

4.4.

Future Work

Overcoming the current hair limitation would help detect potential regional differences in the cerebral response to DCC when comparing regions supplied by different cerebral vasculatures. Due to the limited variability in skin tone in the current study, the analysis focused on hair and not skin tone. A larger and more diverse sample group would allow an analysis of the impact of both hair and skin types on device sensitivity. It would also be useful to conduct studies using tissue-mimicking phantoms to better understand the temporal stability of the Kernel Flow. A further consideration is that most surgeries are performed with the patient supine, which would require testing to ensure the device provides sufficient contact across the head when in this position.