1. Introduction

The oxygen saturation (SO₂) in different brain structures can help clinicians better understand the pathology in various brain diseases and provide insight that may help to improve surgical interventions. There are many different ways of assessing the oxygen level inside the brain where some techniques for example measure the partial oxygen pressure (PtO₂), jugular venous saturation (SjvO₂) or tissue oxygen saturation (StO₂) [1]. However, as SjvO₂ provide assessment of the systemic SO₂ entering the brain, the ability to detect regional ischemia is limited. The PtO₂ method’s limitations are related to the slow response time and the risk of being affected by signal drift. Considering these drawbacks, optical methods provide legitimate alternatives [1, 2].

Optical methods have over the years gained attention as they provide solutions to the need of continuous bedside patient monitoring of physiological parameters such as the oxygen saturation for patients treated for brain injury [3]. Today arterial SO₂ and pulse can be quickly assessed through the widespread use of pulse oximetry [4]. The same principle as used in pulse oximetry and near-infrared spectroscopy (NIRS) based on the difference in absorption between oxygen rich and oxygen deprived Hb can be applied to estimate the local microcirculatory SO₂ with diffuse reflectance spectroscopy (DRS) [5, 6]. The microcirculatory SO₂ is however, not expected to be as high as the arterial SO₂ due to the mix of venules and arterioles that deliver the oxygen to the surrounding tissue. Even though NIRS and DRS are based on absorption of Hb, they use different source detector distances where NIRS has a larger separation in the range of a few centimeters compared to DRS with a corresponding distance of fractions of a millimeter up to a few millimeters. DRS use light produced by a broad spectrum light source in order to sample tissue and estimate the composition of present chromophores. Red blood cells in the microcirculatory blood vessels are packed with Hb that each can carry up to four oxygen molecules and be measured through their specific absorption profile. As the Hb absorbs light differently depending on if it carries oxygen or not, the oxygen saturation can be assessed through spectroscopic analysis.

Previous work involving optical measurements in brain tissue has proved useful for tracking changes in tissue reflectivity related to the anatomy [7]. Several studies have shown that the amount of backscattered light measured by DRS or laser Doppler flowmetry (LDF) can be used to track transitions between different brain tissues such as white and gray matter [8–11]. The differences in backscattered light, known as the total light intensity (TLI) in LDF terminology, could be used for intraoperative guidance during neurosurgical procedures such as DBS surgery. Measurements with patients scheduled for DBS surgery has previously been collected in collaboration with the Department of Neurosurgery at Linköping University Hospital. The LDF data was used to define optical “bar codes” based on the TLI and for studying the microvascular blood flow in front of the optical probe along trajectories toward a DBS target [11, 12]. Deviation in the optical “bar codes” compared to the expected appearance of the TLI signal trend along the preplanned trajectory may be used to indicate a potential brain shift or malfunction of the stereotactic system, which could help to provide safer surgical procedures. In this approach, the probe was inserted with help of a mechanical device adapted to the stereotactic system where the positions of the recordings could be controlled.

A quadratic polynomial fit (QPF) algorithm for estimating SO₂ in human white brain matter from diffuse reflectance spectra has recently been developed for the purpose of real-time monitoring in the neurointensive care [6]. The algorithm uses a few basic chromophores to extract the measured signal. It has been thoroughly evaluated using an experimental model where oxygen saturation could be controlled. In the present work, the developed algorithm is further evaluated on spectra recorded under controlled conditions in the human brain tissue in relation to stereotactic DBS implantation i.e. during creation of the electrode lead trajectory. The aim of the study was to investigate SO₂ and f_B at known anatomical positions along the DBS trajectories and target regions using the QPF algorithm for diffuse reflectance spectra.

2. Material and methods

Patients (n = 11 patients, 7 male and 4 female) ages 43 to 77 (mean ± standard deviation of 66 ± 9 years) referred for DBS implantation at the Department of Neurosurgery, Linköping University Hospital were included in the study. The patients were selected for DBS surgery as treatment for movement disorders (Parkinson’s disease, essential tremor or dystonia) using electrical DBS stimulation. Informed written consent was received prior to surgery and the study was approved by the local ethics committee (M182-04, T54-09). The same patient material has been presented in Wårdell et al., 2016 [11] but in that study the focus was set on high resolution optical recordings of the microvascular blood flow by the use of LDF.

General anesthesia was used for the patients implanted in the DBS target regions of the subthalamic nucleus (STN), the zona incerta (Zi) and the globus pallidus internus (GPi). The surgery was performed partly under local anesthesia and conscious patients when the implantation was done in the ventral intermediate nucleus (Vim) of the thalamus. The surgical procedures followed the protocol for routine DBS-implantation at the clinic. In this concept a stereotactic CT (slice thickness 1 mm, GE Lightspeed Ultra, GE Healthcare, UK) was performed the day of surgery with the Leksell^® Stereotactic System (LSS, Model G, Elekta AB, Sweden). Direct anatomical targeting of the STN, Zi and GPi were done on stereotactic 1.5 T or 3T MRI (T1 and T2, slice thickness 2 mm, Philips Intera, The Netherlands) or after fusion with the stereotactic CT in Leksell Surgiplan^® (Elekta AB, Sweden). The optical measurement probe acts as a guide for creating the trajectory for the DBS lead. When reaching the target area, the probe’s position was verified intra-operatively with fluoroscopy (Philips BV Pulsera, Philips Medical Systems, The Netherlands) and then replaced by the DBS electrode (Lead 3389, Medtronic Corporation, Minneapolis, MN, USA). A postoperative CT was done within 24 hrs. following surgery. A detailed description of the surgical procedure is found in [11].

2.2 Optical measurement setup

The diffuse reflectance spectra ($I_{raw}$) were measured using a spectrometer (AvaSpec 2048-2, Avantes BV, The Netherlands) having a bandwidth of 460-990 nm and resolution of 2.1 nm. Broad spectrum light with wavelengths in the visible and near infrared range (360-2000 nm) was provided by a halogen lamp (AvaLight-Hal-S, Avantes BV, The Netherlands). The light was delivered to the tissue and received using the rigid optic probe with fibers with a core diameter of 125 µm and fiber separation of 250 µm, and a numerical aperture (NA) of 0.37. The optical probe’s outer dimension (length = 190 mm, Ø = 2.2 mm except for the last 30 mm where Ø = 1.5 mm) adapted to the LSS was used. Before surgery, the probe was sterilized according to the STERRAD^® protocol [13].

After each measurement session a white reference spectrum ($I_{white}$) was taken for use in a normalization step where the influence of the light source is compensated, Eq. (1). The white spectrum was measured using a reflective surface, with uniform reflectivity for all wavelengths, of a reference tile (WS-2, Avantes BV, The Netherlands).

2.3 Intraoperative spectroscopic measurement

Spectroscopic measurements were performed along 20 pre-planned stereotactic trajectories ($n_{STN}=8$,$n_{Vim}=6$,$n_{GPi}=4$,$n_{Zi}=2$) with the spectrometer and lamp connected to the probe [11]. The spectra were collected over 30 seconds for each site were the sampling (integration time of 10 ms) was controlled with a LabVIEW interface. The optical probe was inserted with millimeter precision along the pre-planned trajectory with the help of a manually driven mechanical device and the LSS [12]. In a typical trajectory, the optical probe first went through the cortex consisting of gray matter in the outer parts of the brain, past white matter and lighter gray areas in the deep brain, down to the target region. Spectra were recorded in the cortex and at 40, 20 and 10 mm from the target region (Fig. 1). Thereafter measurements were done, in 2.5 mm steps until the pre-calculated target was reached. Along seven STN, two Vim and three GPi trajectories, signals were captured 2.5 mm beyond the target. Further details regarding the surgical procedure are found in [11]. In total, spectra were collected at n = 166 different measurement sites.

 

Fig. 1 a) Deep brain stimulation trajectory toward a target along which spectra were collected, b) axial MRI slice of the STN target region, c) DRS spectrum from gray matter and d) DRS spectrum from light gray matter

Download Full Size | PDF

2.4 Spectral analysis

The QPF algorithm was previously developed [6] and now used for analysis and estimation of the SO₂ and f_B. It uses a second order polynomial fit to the DRS signal in order to assess the ratio between oxygenated and total amount of Hb that corresponds to the oxygen saturation. The shape of the reflected spectrum in the selected region is dependent on the absorption profile of Hb. The quadratic fit, Eq. (2), was applied to a region in the visible spectrum (545-573 nm, Fig. 2) where the absorption of Hb predominates the reflected spectrum. In Eq. (2) the 𝛼, 𝛽 and γ coefficients represent the curvature, x- and y-position, respectively.

 

Fig. 2 Spectral fit for the SO₂ estimation algorithm a) region between 545 and 573 nm (gray-shaded) and normalization wavelength, quadratic fit (marked in red) for b) a HbO₂ dominated spectrum and c) a Hb dominated spectrum

Download Full Size | PDF

The spectral intensity was normalized at 800 nm before the estimations were made. The reduced scattering coefficient was modelled to fit scattering found in human white brain matter. The f_B was calculated using three isobestic wavelengths ($I_{\text{λ}}$) according to Eq. (3) where k and m₁ are constants from the calibration [6].

2.5 Data analysis

Following surgery, the preoperative MRI and postoperative CT images were co-registered and fused in Surgiplan and the final DBS lead position as well as the respective trajectories (Fig. 1) defined though visual inspection by the neurosurgeon (PZ). The Schaltenbrand brain atlas [17] was superimposed when necessary in order to identify anatomical structures. Each final trajectory was investigated and the positions for the optical recordings identified and the tissue type verified. By using this procedure, the influence from potential brain shift was reduced. A detailed presentation of the method for anatomical identification can be found in [12].

The spectra were divided into groups based on the anatomical position of the recordings according to: gray matter (n = 8 measurement sites), white matter (n = 36), and deep brain tissue referred to as light gray matter (n = 122). The recordings in the target regions were included in the light gray matter group. A separate analysis was performed with measurements in the target groups STN, Vim and GPi. As only two spectra were collected in the Zi region these were not included in the target group analysis. Spectra recorded a few mm beyond the targets (approx. 2.5 mm) were included in the respective target group. Median values of the SO₂ and f_B and the light intensity at 780 nm normalized to the highest value in white matter for each patient (I_norm,780) were calculated for the respective groups. Box plots show median values and quartiles (with Q1 (25%) to Q3 (75%) percentiles with box) for each data set. The statistical analysis was performed with Minitab^® (Minitab Inc., UK). Test for normal distribution (Anderson-Darling) indicated that the data was not normally distributed and therefore double-sided nonparametric Mann-Whitney tests were used to compare medians between groups. A p-value less than 0.05 were considered significant.

3. Results

All DBS leads were implanted in the pre-planned target regions except for two trajectories, where one towards Zi resulted in a displacement 6 mm medial to the target, and a second trajectory had a deviation of 1.5 mm, both cases were without surgical complications such as bleeding or the need for reoperation. The optical DRS measurements did not extend the surgical procedure more than 5 minutes per trajectory.

Figure 3 shows example of spectra from gray, white and light gray matter. Calculated SO₂ and f_B are shown for each spectrum and the quadratic fits are marked with a dotted red line. In Fig. 3(a), a clear typical double peak of hemoglobin is found for gray matter whereas lower SO₂ and f_B were found in the white and light gray matter.

 

Fig. 3 Examples of spectra along the DBS trajectories in a) gray matter with high SO₂ and f_B, b) white matter with lower SO₂ and f_B and c) light gray matter with low SO₂ and f_B

Download Full Size | PDF

The median SO₂ as estimated for white, light gray and gray tissue groups are seen in Fig. 4(a) with median values of 52% (n = 8) for gray matter, 24% (n = 36) for white and 20% (n = 122) for light gray matter. The median SO₂ was different between light gray and white tissue (p < 0.001) and between light gray and gray tissue (p < 0.005). The difference in median SO₂ between white and gray tissue was not significant (p = 0.08 > 0.05) for these limited number of measurement sites. The median f_B in Fig. 4(b), for gray matter (3.9%) was significantly different from white (1.0%, p < 0.05) and light gray matter (0.9%, p < 0.001) while no difference was found between the two latter groups. The normalized light intensity at 780 nm (I_norm,780) in Fig. 4(c), was significantly different (p < 0.001) between the tissue groups with medians of 0.90 for white matter, and 0.67 for light gray tissue and 0.38 for gray matter.

 

Fig. 4 Boxplots of estimated values in gray, white and light gray brain tissue for a) oxygen saturation (SO₂), b) blood fraction (f_B) and c) normalized light intensity at 780 nm (I_norm,780)

Download Full Size | PDF

The median SO₂ for the target regions were 20% (n = 16) for STN, 16% (n = 11) for Vim and 21.4% (n = 7) for GPi. Estimation of f_B for the analyzed target regions had medians of 0.9% for STN, 0.9% for Vim and 1.2% for GPi. No significant differences in SO₂ or f_B were found between target regions, Figs. 5(a) and 5(b). The I_norm,780 seen in Fig. 5(c) resulted in median values of 0.81 for STN, 0.60 for Vim and 0.66 for GPi. Differences in median I_norm,780 were found between the analyzed target regions, STN and Vim (p < 0.001), STN and GPi (p < 0.005) and between Vim and GPi (p < 0.05).

 

Fig. 5 Boxplots of the estimated parameters for DBS targets: STN, Vim and GPi with a) oxygen saturation (SO₂) and b) blood fraction (f_B) and c) normalized light intensity at 780 nm (I_norm,780)

Download Full Size | PDF

The median SO₂ within the measurements for each patient are presented in Fig. 6(a) were the patients had an overall median SO₂ of 24% and ranged between 17 to 26%. The median f_B for each patient varied between 0.6 and 2.3% while the median I_norm,780 for individual patients was calculated and ranged between 0.57 and 0.83, as seen in Figs. 6(c) and 6(d).

 

Fig. 6 Boxplots of the estimated parameters for DBS patients 1 to 11 for a) oxygen saturation (SO₂) and b) blood fraction (f_B) and c) normalized light intensity at 780 nm (I_norm,780)

Download Full Size | PDF

4. Discussion

In this study a novel algorithm was used to estimate the oxygen saturation and blood fraction along trajectories towards targets in the deeper part of the brain in eleven patients undergoing DBS-surgery. An algorithm based on QPF to the measured signal was for the first time used on a larger set of human brain tissue spectra captured from well-defined anatomical positions. This was accomplished by a DRS system adapted for use with the Leksell^® Stereotactic System, a manually controlled mechanical device and the application of the algorithm on the recorded spectra. It was found that the SO₂ and f_B was different in gray matter compared with light gray and white matter whereas no significant difference was found in SO₂ and f_B between target regions.

The SO₂ estimations using the suggested algorithm produced values of approximately 50% for cortical and superficial gray matter and 20-30% in deeper regions of the brain. In a study by Johns et al. the authors describe a method for analyzing light in the visible region based on modified diffusion theory to estimate the SO₂ [18]. The authors presented estimations from spectroscopic data in human brain with SO₂ of 54% in gray matter (4 mm deep) and 26% in white matter (33 mm deep) [18]. These values correspond well with the median SO₂ values for gray and white matter presented in Fig. 4 from 11 patients and in total 166 recordings. The authors in [18] also state that the accuracy of the estimation method is better for well-perfused tissue with a hemoglobin concentration over 50 µM which roughly corresponds to a f_B over 2% for a normal hemoglobin concentration of 2.4 mM in the blood. Similarly, an adequate Hb concentration which result in clear spectral features, is also needed in order to produce reliable estimations for the algorithm presented by Rejmstad et al. [6]. For the latter method, optical phantoms with scattering properties corresponding to white brain matter and f_B between 0.5 and 3% were used for calibration [6]. As the scattering in white and gray brain matter are different, the effects of these differences for the SO₂ estimation were investigated. Large variations in the reduced scattering resulted in errors of up to 4% points for SO₂ between 0 and 50% and up to 12% points for SO₂ at 100% [6].

Other studies measuring SO₂ in the brain have for example used NIRS through the intact skull, detecting changes in cortical blood flow where high SO₂ values are expected. A review by Epstein et al. discussed the clinical use of the NIRS method for tissue oxygen saturation monitoring where StO₂ values lower than 70-75% were associated with increased risk of organ failure and mortality [19]. In a recent study by Sommer et al. a commercial device based on LDF and DRS known as “Oxygen-to-see” or O2C was used on 20 SAH patients reporting baseline SO₂ of 39%, in subdural (7 mm deep tissue) brain tissue [20]. A translation from measured intraparenchymal pO₂ (20 - 35 mmHg) [21] to SO₂ via the oxygen dissociation curve [21], using reported baseline values for CO₂ = 50 mmHg, pH = 7.15 and T = 37 °C [22], these conditions corresponds to an SO₂ between approximately 30 and 60% which is higher than the acquired estimations in this work. Possible reasons for the difference compared with the reported numbers could be related but not limited to a low amount of Hb in the collected spectra, deviations in physiological parameters such as pH and CO₂ compared to the “normal” modelled conditions, scattering variation effects as the model was based on scattering in white matter or influence from other potential chromophores not included in the estimation model.

In a similar study using DRS by Johansson et al. corresponding f_B ranged from 0.2% to 1% with a median of 1% for cortical tissue and 0.2% for STN compared to the values of 3.9% and 0.9% for gray matter and STN in this study [10]. A possible reason for the differences is for example that the study by Johansson et al. included additional chromophores such as Lipofuscin and cytochromes in the estimation algorithm [23]. For the same algorithm as used in this study, adapted to a fiber separation of 300 µm instead of 250 µm, experimental phantom measurements were used for calibrating the estimations for f_B values between 0.5% and 3% [6].

In addition to the DRS measurements, LDF was used to study the microvascular blood flow using light with a wavelength of 780 nm [11]. The intensity of the reflected light from DRS at the same wavelength was used in order to enable comparison between the two modalities. After normalizing the reflected intensity at 780 nm to the intensity in white matter for each patient the relative intensity (I_norm,780) in the DBS target regions, Fig. 5(c), resulted in median I_norm,780 in agreement with the differences between white matter and STN (approximately 20%) and between white matter and GPi (approximately 36%) reported by Antonsson et al. [8]. The I_norm,780 values for the Vim and STN targets also showed agreement with a previous report presenting values of the TLI using LDF. In the study by Wårdell et al. [12], TLI values in the target regions were normalized to white matter and were 0.6 ± 0.1 for Vim and 0.9 ± 0.08 for STN comparable with I_norm,780 of 0.66 for Vim and 0.8 for STN in the current study.

Monitoring of oxygen saturation in brain tissue is of great importance in neurointensive care and can be used to detect hypoxia for patients at risk. Medical conditions that would benefit from oxygen monitoring are for example patients with traumatic brain injury and subarachnoid hemorrhage [24, 25]. A benefit of using optical methods such as DRS is the possibility to perform multimodal, simultaneous measurements by also using other optical techniques such as LDF. This can be accomplished by using a single optical probe for both systems, but with a flexible tip easily inserted and fixated in the brain tissue [26]. By using techniques that support continuous collection of signals with high temporal resolution, trends of changes over time can be of use to clinicians during patient monitoring [25]. A future step could be to use DRS and the suggested algorithm for real-time estimation of SO₂ and f_B in combination with LDF to gain continuous and bedside measurements of the microvascular circulation. A benefit of acquiring information for both blood flow and oxygenation is that changes in the relative metabolic rate of oxygen could be tracked in different tissues or over longer periods in the neurointensive care.

In summary: SO₂ and f_B was estimated using a novel algorithm from diffuse reflectance spectra along stereotactic trajectories collected during DBS surgery. The median SO₂ estimated for measurements in gray, white and light gray matter were 52%, 24% and 20%, respectively, using the suggested algorithm. A major advantage is the algorithms capability to present real-time information useful in clinical neurointensive care monitoring.