Cerebral time domain-NIRS: reproducibility analysis, optical properties, hemoglobin species and tissue oxygen saturation in a cohort of adult subjects

: The reproducibility of cerebral time-domain near-infrared spectroscopy (TD-NIRS) has not been investigated so far. Besides, reference intervals of cerebral optical properties, of absolute concentrations of deoxygenated-hemoglobin (HbR), oxygenated-hemoglobin (HbO), total hemoglobin (HbT) and tissue oxygen saturation (StO 2 ) and their variability have not been reported. We have addressed these issues on a sample of 88 adult healthy subjects. TD-NIRS measurements at 690, 785, 830 nm were fitted with the diffusion model for semi-infinite homogenous media. Reproducibility, performed on 3 measurements at 5 minutes intervals, ranges from 1.8 to 6.9% for each of the hemoglobin species. The mean ± SD global values of HbR, HbO, HbT, StO 2 are respectively 24 ± 7 μ M, 33.3 ± 9.5 μ M, 57.4 ± 15.8 μ M, 58 ± 4.2%. StO 2 displays the narrowest range of variability across brain regions.


Introduction
Near-infrared spectroscopy (NIRS) is a non-invasive methodology able to provide continuous monitoring of cerebral oxygenation parameters at the bed-side of the patient. Despite over 35 years have passed since its first description [1,2], NIRS application has been hindered in the routine clinical settings by a number of limiting factors (e.g. lack of absolute data, confounding effects due to systemic superficial changes or movement artifacts) [3][4][5]. Some limitations have been overcome thanks to the technological advances. In particular, differently from the classical continuous-wave technology (CW-NIRS) [6], the frequency-(FD-) and time-domain (TD-) NIRS [7] allow to calculate the absolute absorption and reduced scattering coefficients, to recover the absolute concentrations of hemoglobin species and to better separate the contribution of the extra-cerebral layers from the brain cortical ones.
The first steps to apply efficiently NIRS in neurology clinic should be to evaluate the in vivo reproducibility of the technique, to define normal values of cerebral deoxy-hemoglobin (HbR), oxy-hemoglobin (HbO), total hemoglobin (HbT) and tissue oxygen saturation (StO 2 ) and to assess the variability among brain areas and between hemispheres. Previous studies analyzed optical properties of adult head in vivo with FD-or TD-NIRS techniques on a limited number of subjects [8-16] registered exclusively from the forehead except from one study [14]. The reported ranges of normal values for HbR, HbO, HbT concentrations and StO 2 were heterogeneous (Table 1).
Reproducibility, which is defined as the variation in measurements under changing conditions such as time [17,18], has been tested in vivo in few studies, carried out mainly with the CW-NIRS technique (for StO 2 ) in neonates, with contrasting results [19][20][21][22][23][24]. The reproducibility of FD-NIRS in adult subjects was investigated in one recent study that reported correlation coefficients between measurements taken 5 months apart ranging from 0.7 to 0.9 for the hemoglobin species [11].
The aim of the present study is to assess reproducibility of TD-NIRS measurements under changing conditions of time and after removing and repositioning of the optical fibers (shortterm reproducibility). In addition, we describe the variability between hemispheres and head regions and the range of normal values of TD-NIRS measurements obtained from three different head regions (frontal, central, and parietal) in a relatively large cohort of adult healthy subjects.

Subjects
The study received approval of the Ethical Committee of San Raffaele Scientific Institute and the Italian Ministry of Health and was conducted in compliance with the Declaration of Helsinki. Each subject gave written informed consent. Adult healthy subjects [25] were enrolled according to the following inclusion criteria: age ≥18 years old; absence of stenosis ≥ 50% in cervical extra-cranial and intra-cranial arteries according to Doppler ultrasonography; absence on head CT-or MRI-scan, when available, of high-grade cerebral small vessel vasculopathy (Fazekas scale > 2) or other relevant pathologies.

Instrumentation and data acquisition
The , obtained facing the injection and collection fibers with a thin Teflon film in between to properly excite all modes in the fiber bundle, has a full width at half maximum (FWHM) of <200 ps. Figure 1 shows an example of IRF and of TD-NIRS data on a subject.

Measurement protocol
TD-NIRS measurements were obtained from frontal, central and parietal brain regions on each hemisphere. The optical probes were positioned by means of a modified EEG cap (g.EEGcap, g.tec medical engineering GmbH, Austria) according to the 10-10 international system (position F3-F5, C3-C1, P3-P5 for the left hemisphere and C4-C2, F4-F6, P4-P6 for the right hemisphere) and held manually to improve stability (Fig. 2). According to probabilistic anatomical correlation of 10-10 international system to the brain cortex  The injected power of each position was optimized to collect at least 10 6 photons for each wavelength in a maximum acquisition time of 30 s.
The study protocol was performed in a quiet room, in condition of low light with subject at rest laying with angle-of-bed inclination at 30°. Reproducibility of measurements was tested by performing 3 measurements at 5 minutes intervals after removing and repositioning of the optical probes (short-term reproducibility).
During the experimental session of each subject, the IRF and the TD-NIRS curves from a solid phantom [nominal optical properties at 660 nm: absorption (μ a ) 0.1 cm −1 and reduced scattering coefficient (μ s ') 10 cm −1 ] were measured at least once to monitor the performances of the device. Therefore, the quality of phantom was established as the mean ± 2 standard deviations (SD) of the distribution of fitted optical properties of all measurements. Quality assessment included Δt TDNIRS-IRF , the difference between the temporal position of the barycenter of TD-NIRS photon curves and the barycenter of the IRF. We then calculated the mean ± 2SD of the distribution of Δt TDNIRS-IRF for every pairs of TD-NIRS and IRF curve of the entire cohort of subjects. Then we discarded TD-NIRS curves having a Δt TDNIRS-IRF larger than 2 SD of the mean time shift (see paragraph 2.4).

Data analysis
TD-NIRS data at each wavelength were fitted with the solution of the diffusion equation for semi-infinite homogenous media after convolution with the IRF [31]. A spectral constraint during the inversion procedure was employed [32]. In particular, we used the empirical spectral dependence for the reduced scattering (obtained by Mie theory [33,34]): where λ is the wavelength, λ 0 is a reference wavelength (600 nm in this case), a is the estimated reduced scattering coefficient at λ 0 , b is the so called scattering power. Note that a and b can be related to the equivalent density and size of scatters [33,34]. Further, we used the Beer's law to express the absorption coefficient as a linear superposition of the specific absorption of HbR, HbO and water: (2) Water content was fixed at 80% while HbR, HbO, a, and b where used as free fitting parameters. The source detector separations corresponded to the distances between the holes of the EEG cap that measured for each size of the EEG cap (S, M and L). The fitting range included all points with a number of counts higher than 90% of the peak value on the rising edge of the curve and 1% on the tail. A Levenberg-Marquardt procedure was employed to minimize the least square error between data and model as estimated by the reduced chisquare (χ 2 ). For comparison with literature data we calculated μ a (λ) and μ s '(λ) from the fitted values of HbO and HbR and from a and b, respectively using Eqs. (1) and (2).
The following quality control parameters were applied to each measurement: number of detected photons >20,000 (for at least 2 out of 3 wavelengths); χ 2 ≤10; phantom's µ a or µ s ' values not exceeding the mean ± 2SD of all phantom measurements; Δt TDNIRS-IRF not exceeding mean ± 2SD of the distribution of Δt TDNIRS-IRF in our cohort.
One-way repeated-measure ANOVA was used to evaluate reproducibility considering the three head areas of right and left hemisphere. According to Bonferroni correction, we set the significance level cut-off at p<0.008. The square root of residual mean square obtained from one-way repeated-measure ANOVA corresponds to the pooled within-subject standard deviation (SD w ), which represents the intra-subject variation in the measurements [17,35]. The Bland-Altman (95%) repeatability limit, which is the expected maximum difference between two measurements on the same subject in 95% pairs of observations, was calculated according to the following equation [17,35]: where RL is the repeatability limit, SD w is the within-subject standard deviation calculated as described before. The RL was proposed by Bland and Altman as a simple way to present measurement errors in reproducibility studies and has been used in previous studies [17,35,36]. Data analyses were performed with SPSS Statistics for Macintosh, version 20.0 (Armonk: IBM Corp.).

Reproducibility analysis
Among the three recording sessions of each subject HbR, HbO, HbT and StO 2 values obtained from each position did not show significant intra-subject difference ( Table 3). The SD w (within-subject variation) was ≤4.1% for StO 2 and ≤7% for HbR, HbO, HbT (Table 3). The intra-class correlation coefficients (ICC) demonstrated an excellent reliability of repeated measurements for µ a (690 nm), µ a (785 nm), µ a (830 nm), HbR, HbO, HbT and StO 2 variables (ICC>0.8) in each position (Tables 3, 4). The ICCs for µ s '(690 nm), µ s '(785 nm), µ s '(830 nm) showed instead a fair to good reliability (ICC 0.5-0.7) ( Table 5). The Bland-Altman plots of the TD-NIRS values of the different hemoglobin species indicate a small interval of variation between the three recordings (Fig. 3). After quality control the SD w (within-subject variation) improved from 8.9% to 4.4% for HbR, from 11.1% to 5.6% for HbO, from 7.5% to 3.8% for HbT, from 7.3% to 2.7% for StO 2 .

Discussion
To the best of our knowledge, this is the first study on a large cohort of healthy adult subjects that assessed TD-NIRS short-term reproducibility of absolute values of HbR, HbO, HbT, StO 2 , µ a and µ s '. Optical properties and hemoglobin parameters displayed non-significant differences among repetitions, with StO 2 showing the lowest within-subject variation ranging from 1.8% to 4.1%. While large studies have reported reference values of NIRS measurements in the newborns [37], few data are available in adults (Table 1). We report optical properties (μ a , μ s ') and hemoglobin species (HbR, HbO, HbT, StO 2 ) obtained with TD-NIRS from different head regions of adult subjects.
We acknowledge the use of a homogenous model instead of a two-layer algorithm as a possible limitation of our study. The reason for using a homogeneous model stems from the observation that in a previous multi-laboratory study [16], a substantial agreement was found, both in simulations and in vivo measurements, between TD-NIRS absorption coefficients retrieved from the homogenous model and the estimate of the absorption coefficient in the lower layer derived by exploiting a two-layer model.
Our results compare to TD-NIRS data by Auger et al. . These results might appear unexpected considering previous studies reporting a StO 2 of 44-69% from jugular bulb venous oximetry in adults [38,39] and an estimated arterial fraction of cerebral blood volume of 30% from PET studies [40]. Substantial differences in the object of the investigation between the two techniques might explain this apparent discrepancy. Indeed, while the jugular venous system drains venous blood from the whole brain tissue (white and gray matter as well as superficial and deep brain structures), TD-NIRS investigates almost only the microcirculation of the grey matter. Since the metabolism of cerebral cortex is roughly two to three times that of the white matter [41,42], cortical venous blood may display lower StO 2 due to higher oxygen extraction compared to other brain regions. On the other hand, the potential impact of the superior sagittal sinus in lowering StO 2 should be negligible according to the following considerations: the main signal of TD-NIRS should come from the microcirculation since large venous structures would hugely absorb injected photons; the distance of ≈3 cm from midline position (Cz) to C1 or C2; the central regions exhibited values of StO 2 comparable to the other head regions. Nevertheless, we observed significant differences in hemoglobin species concentrations among different head regions. In a speculative way, we can hypothesize that these differences might be due to variability in cerebral blood volume, cerebral blood flow and metabolism among different head areas. According to previous PET and MRI studies, a default mode network regulates the activation and deactivation of cerebral regions in resting conditions [43]. The P3-P5 and P4-P6 areas should in part involve the Broadman area 19 [29], which deactivates in resting conditions in the context of default mode network [43]. Parietal regions also exhibited significantly reduced concentrations of HbR and HbO respect to frontal and central regions that were more similar for concentrations of HbR and HbO. The inter-subject variability for the hemoglobin species found in our study cohort needs confirmation in future studies with a larger number of subjects.
Previous studies assessed the reproducibility of NIRS on newborns, mostly using CW-NIRS, with contrasting results (2.4-14.8% for StO 2 ) [19-24]. However, comparison with our study is limited considering the different type of subjects and instrumentations. The only study that assessed the reproducibility of FD-NIRS on adult subjects was performed by Hallacoglu et al. [11]. They found reproducible results according to the reported cross correlation coefficients between measurements taken 5 months apart of ∼0.9 for μ a (690 nm), μ a (830 nm), HbR, HbO, HbT, ∼0.8 for StO 2 , ∼0.7 and ∼0.6 for μ s '(690 nm) and μ s '(830 nm) respectively. We tested reproducibility of TD-NIRS measurements in different head regions at short-term, 5 minutes intervals, after removing and repositioning optical fibers. Indeed, NIRS measurements, performed after a short time period, could be influenced by physiological variations in cortical activation. In this regard, previous fMRI, fMRI-NIRS and fMRI-EEG studies have demonstrated that spontaneous low frequency oscillations (<0.1 Hz) do occur at rest and could be related to changes in blood oxygenation and blood flow [43][44][45][46][47]. Besides, it is recognized that cognitive activity is still present in resting condition and even during sleep [47].
Despite differences in study protocol, we found ICC coefficients ranging from 0.95 to 0.99 for μ a (690 nm), μ a (785 nm), μ a (830 nm), HbR, HbO, HbT and from 0.8 to 0.94 for StO 2 similar to the study of Hallacoglu and colleagues [11]. Compared to previous studies, we tested more head regions, which might increase the consistency of our results.
Our results suggest that HbR, HbO, HbT and, to a lesser extent, StO 2 values reduce with increasing age. Since we are investigating only a limited portion of the brain, in particular the cerebral cortex of six brain regions, we cannot infer on the modifications of global metabolic and hemodynamic cerebral properties that occur with aging. Nevertheless, we know that a number of confounding factors, including atrophy, might become relevant to cerebral metabolism with aging. Indeed, other imaging modalities, such as PET or MRI, reported a progressive reduction of cerebral metabolic rate of oxygen with increasing age without concomitant modifications of oxygen extraction fraction [41,[48][49][50][51][52][53]. Moreover, the effect of age on cerebral blood flow and cerebral blood volume is controversial as it has been shown that increasing atrophy might be a significant confounding factor [49,51,[54][55][56][57]. The reduction in HbT may suggest either a decrease in cerebral blood volume and flow, as well as an effect of increasing atrophy [58][59][60], hemodilution [61] or flow redistribution. In our study the decline of HbR, HbO, HbT and, to a lesser extent, StO 2 appeared more evident after the 5th decade. Although this is an exploratory cut-off, which needs further confirmation, from a biological point of view it appears appropriate to account for aging. TD-NIRS values of cerebral hemispheres and across brain's regions will be instrumental in planning future studies on neurological diseases, such as acute ischemic stroke, because different brain areas are involved in any given patient.
One strength of our study is the selection criteria of subjects. We gathered medical history and performed neurological examination to exclude relevant neurological comorbidities. In addition, each subject underwent Doppler ultrasonography screening to exclude individuals with ≥50% stenosis of extra-and intra-cranial arteries. We also reviewed available neuroimaging, CT or MRI head scans, which allowed us to exclude significant intra-cranial pathologies. Twenty-six subjects without neuroimaging displayed comparable HbR, HbO, HbT, and StO 2 mean values compared to the remaining sample [respectively, p = 0.78; p = 0.44; p = 0.56; p = 0.31)].
Quality control criteria prevented the inclusion of unreliable measurements. Indeed, we have shown that reproducibility improved after the application of quality control criteria. It is worth mentioning that discard of measurement was related mostly to malfunctioning of the device (quality controls for Δt TDNIRS-IRF and phantom measurements) or to movement artifacts (quality controls for Δt TDNIRS-IRF and χ 2 >10) and seldom to low photon counts or hair contamination. The application of these quality control criteria might standardize the measurements of future studies and increase the reproducibility among laboratories.

Conclusion
In summary, we have provided evidence that TD-NIRS brain measurements are reproducible and we have reported ranges of normal values for absorption and reduced scattering coefficients, oxy-, deoxy-and total hemoglobin concentrations and oxygen saturation in a large cohort of adult subjects. StO 2 displayed the narrowest range of variability across brain regions. Aging appeared to have an impact on hemoglobin species. The results of this work are pivotal for the application of absolute concentrations assessed by TD-NIRS in research and clinical field of neurological diseases, such as ischemic stroke, neurodegenerative diseases, migraine and epilepsy.