Intraparenchymal near-infrared spectroscopy for detection of delayed cerebral ischemia in poor-grade aneurysmal subarachnoid hemorrhage

Objective: Detection of delayed cerebral ischemia (DCI) is challenging in comatose patients with poor-grade aneurysmal subarachnoid hemorrhage (aSAH). Brain tissue oxygen pressure (PbtO2) monitoring may allow early detection of its occurrence. Recently, a probe for combined measurement of intracranial pressure (ICP) and intraparenchymal near-infrared spectroscopy (NIRS) has become available. In this pilot study, the parameters PbtO2, Hboxy, Hbdeoxy, Hbtotal and rSO2 were measured in parallel and evaluated for their potential to detect perfusion deficits or cerebral infarction


Introduction
Despite steady progress in the treatment of aneurysmal subarachnoid hemorrhage (aSAH), high mortality rate and poor functional outcome persist in severely affected patients (Nieuwkamp et al., 2009).One of the main reasons is the occurrence of delayed cerebral ischemia (DCI) in 20-30 % of patients (Galea et al., 2017).Vasospasm of large cerebral vessels plays a crucial role in the occurrence of secondary ischemia.However, other mechanisms have been discovered in recent years, such as microthrombi, impaired microcirculation, neuroinflammation, disruption of the blood-brain barrier, defect cerebral autoregulation and cortical spreading depolarization (Balbi et al., 2017;Hurth et al., 2020;Tso and Macdonald, 2014).It remains unpredictable which patient will develop DCI and when.Additionally, it is important to distinguish between DCI and delayed cerebral infarction.Whereas the former describes acute cerebral hypoxia (defined as clinical deterioration in terms of a new focal neurological deficit or a decrease of at least 2 points on the Glasgow Coma Scale lasting for more than 1 hour (Vergouwen et al., 2010) and might be reversible, the latter describes an irreversible infarction confirmed by neuroimaging.Detection of cerebral hypoxia is a major challenge, especially in comatose patients (Rass and Helbok, 2021).In these patients, multimodal neuromonitoring using brain tissue probes allows to continuously monitor cerebral oxygenation and cellular metabolism.By measuring partial brain oxygen pressure (PbtO2), cerebral hypoperfusion might be detected and thus treated earlier which is associated with fewer infarcts and better prognosis (Veldeman et al., 2020).Because PbtO2 monitoring is an invasive procedure, alternative non-invasive measurements of cerebral tissue oxygen saturation have already been investigated.Conventional near-infrared spectroscopy (NIRS) is a non-invasive technique that uses light in the near-infrared range to measure the relative proportion of oxyhemoglobin (Hboxy) and deoxyhemoglobin (Hbdeoxy) in brain tissue.The calculated total hemoglobin (Hbtotal) and regional oxygen saturation (rSO2) reflect changes in oxygen metabolism.Only few studies applying NIRS with plaster-based patches attached to the skin, have investigated and compared the two measurement methods (PbtO2 and rSO2) in parallel (Davies et al., 2019;de Courson et al., 2022;Roldán and Kyriacou, 2021;Rosenthal et al., 2014).The results are controversial, as extracranial tissue perfusion and the skull can affect the measurement and lead to inaccurate readings of non-invasive NIRS measurements (Sørensen et al., 2015).Recently, a new brain tissue probe for combined measurement of intracranial pressure (ICP) and intraparenchymal near-infrared spectroscopy (NIRS) has become available (Seule et al., 2016a(Seule et al., , 2016b;;Keller et al., 2011).It allows to collect NIRS-derived parameters directly in the brain tissue, bypassing the interfering factors.
In this pilot study, we aimed to evaluate the potential of the NIRS-ICP probe in detecting cerebral hypoperfusion and/or infarction.In contrast to prior studies, (Keller et al., 2011;Seule et al., 2016b) we focused on collecting data on Hboxy, Hbdeoxy and Hbtotal rather than cerebral blood flow (CBF).Furthermore, we compared our results with PbtO2 measurements collected simultaneously.

Materials and methods
In this prospective cohort study data from the Neurocritical Care Units University Hospital Zurich and Charité Universitätsmedizin Berlin were analyzed.Zurich data was collected between December 2019 and March 2022, Berlin data between September 2020 and January 2022.
The study was approved by the local ethics committees (Kantonale Ethikkomission, BASEC-Nr.2016-01101; Bioethics Committee of the Medical University of Charité -University Medicine Berlin (EA4/033/ 20)).At both hospitals written consent was given by legal representatives, as all patients were incapable of judgment.The data supporting this study are available from the corresponding author upon reasonable request.

Study population and treatment
In patients with poor-grade aneurysmal SAH (aSAH) who remained comatose or had to be sedated after successful aneurysm treatment, extended neuromonitoring using PbtO2 (Licox, Integra LifeScience, Plainsboro, USA), NIRS-ICP probe (RheoSens, RheoControl, RheoLity, Luciole Medical, Zurich, Switzerland) and microdialysis (Iscus flex , M Dialysis AB, Stockholm, Sweden) were installed.Poor-grade aSAH was defined as WFNS grade ≥ IV and/or a modified Fisher grade ≥ 3 at admission.All patients were analgosedated and mechanically ventilated.The probes were placed via a common 3-lumen bolt (IM3_EU, Licox, Integra LifeScience, Plainsboro, USA).For obtaining information from two vascular territories (middle and anterior cerebral arteries), the probes were placed via a borehole in the frontal watershed regions.In Berlin, patients were monitored only unilaterally and neuromonitoring was placed on the side of the ruptured aneurysm or, in the case of midline aneurysm, on the side with the most subarachnoid blood.In Zurich, a bilateral placement was performed whenever possible.However, this was avoided in patients requiring hemicraniectomy or with other neurosurgical contraindications.
Patients were treated according to the guidelines of the Neurocritical Care Society (Diringer et al., 2011) and the respective standard therapies of the two centers regarding aSAH.Common to both and to be emphasized in brief is the goal of keeping the ICP below 20 mmHg, the CPP above 60 mmHg and the continuous intravenous or oral administration of nimodipine.In addition, daily transcranial Doppler sonographies were carried out whenever possible.
No regular or scheduled CT perfusion imaging was performed for the possible detection of DCI.CT perfusion imaging was only performed if there were indications of a possible cerebral perfusion disorder.A decrease in PbtO 2 below 20 mmHg for more than 30 mins and/or an increase in the lactate/pyruvate ratio above 40 or a significant increase in the blood flow velocities determined by transcranial Doppler sonography (TCD) mean blood flow velocities in the middle cerebral arteries >120 cm/s were used to screen patients eligible for perfusion CT.No therapeutic or diagnostic decisions were made based on NIRS measurements.
In CT perfusion imaging, we confirmed a DCI by delaying the Tmax over 6 seconds in the respective area.Tmax is a value calculated by the RAPID software and a sensitive parameter to identify critically hypoperfused penumbral tissues (Dong et al., 2019).
DCI treatment was performed according to in-house protocols, which did not differ significantly between the two centres.Initially, euvolemic arterial hypertension was induced and maintained with optimization of O 2 delivery.If the multimodal monitoring values remained low, endovascular therapy in the form of intra-arterial spasmolysis or angioplasty was performed after further confirmation by imaging.

The NIRS-ICP probe
The NIRS-ICP probe is a parenchymal probe with sensors for intracranial pressure (ICP) and temperature that is additionally supplied with optical fibers and optical sensors for NIRS measurements (RheoSens, RheoControl, RheoLity, Luciole Medical, Zurich, Switzerland).It provides continuous measurements of Hboxy and Hbdeoxy.The Hbtotal is given by the sum of the two and the oxygen saturation rSO2 as the ratio Hboxy divided by Hbtotal.

Hb total = Hb oxy + Hb deoxy rSO2 = Hb oxy Hb total
The probe additionally offers the potential for quantitative assessment of CBF and cerebral blood volume (CBV) through a combined NIRS and ICG dilution mode, (Seule et al., 2016b) which was not part of this study.For implantation and placement of the three probes (NIRS, PbtO2 and microdialysis), a bolt (IM3_EU, Licox, Integra LifeScience, Plainsboro, USA) was inserted through a frontal burr hole.The positioning of the probes was verified by a postinterventional CT scan.

Outcome definition
We aimed to determine to which extent the readings from the PbtO2 and NIRS probes allow to predict cerebral hypoperfusion and/or infarction events.Initially, suspicion of perfusion deficit was based on multimodal neuromonitoring: PbtO2 <20 mmHg, TCD mean blood flow velocities >120 cm/s or increase by 50 cm/s within 24 h, cerebral microdialysis (CMD) with lactate-pyruvate ratio >40 (Rass and Helbok, 2021).The two endpoints were defined as: 1. Perfusion deficit determined by contrast enhanced computed tomography (CT) (Tmax > 6 s in at least 1 vascular territory).2. Occurrence of a new infarction unrelated to surgery and identifiable in CT images.The imaging data was assessed by a trained neurologist blinded to the probe data at the time of assessment.

Data processing and analysis
At both centers, the PbtO2 values were measured by Clark type electrodes (Licox, Integra LifeScience, Plainsboro, USA) and the NIRS data (rSO2, Hboxy, Hbdeoxy, Hbtotal) by the Luciole system (RheoSens, RheoControl, RheoLity, Luciole Medical, Zurich, Switzerland).Data were collected and time-synchronized via the Moberg Component Neuromonitoring Systems (CNS) (Moberg Research Inc, PA, USA).In Zurich, the Moberg CNS devices are integrated into the ICU Cockpit IT platform (Boss et al., 2022) which automatically writes collected time series data into a SQL research database.The data for the analysis were exported from the SQL research database.In Berlin, the data were exported for each patient directly from the Moberg CNS.In patients with bilateral placement of probes, the data from the individual hemispheres were assumed to be independent.
The time-aligned data with sampling frequencies of 1 Hz and 25 Hz for the PbtO2 and NIRS signals, respectively, were subsequently downsampled to a sampling time of 10 seconds using a median filter.A reviewer, blinded to the outcomes, labeled obvious artifacts in the preprocessed data as such using a graphical labeling tool (Label Studio v1.4.1, https://labelstud.io/).Times labeled as artifacts were subsequently indicated as missing.
In the resulting cleaned data, the short time correlation between PbtO2 and rSO2 was assessed for 15 min frames for which both signals were simultaneously available.Likewise, the correlation of the concatenated median values for these 15 min frames of all patients was calculated to determine the overall correlation in the signal levels over longer timescales.
We then located the time points when imaging via CT or MRI was performed and for which the outcome was assessed.For the two time frames of 24 and 48 h immediately before imaging events, we computed median values and hypoxia burden.Hypoxia burden was thereby defined as the area between the curve below the threshold and the threshold itself.Importantly, as the hypoxia burden critically depends on the chosen threshold, we computed the hypoxia burden for each parameter of a range of thresholds and studied which threshold would exhibit the strongest association with the outcome.If a time frame had data for less than 10 % of its time points for any signal, it was excluded.In patients with a bilateral placement, two sets of median values and hypoxia burden were derived and compared with the findings in imaging events of the corresponding hemisphere.To this end, all hemispheres were weighted equally.The 24 and 48 h time frames were chosen to reflect the clinical significance of an early detection of an imminent complication as well as a worsening of the patient condition, respectively.

Statistical analysis
We plotted boxplots and violin plots for the median and burden extracted from the different signals and thresholds in Figs.1-2.Mann-Whitney-U test was used for comparing continuous variables.Statistical significance was assumed at P < 0.05 using a two-sided alternative.The effect size was quantified using the ROC AUC value where a value above 0.5 indicates a positive association and below 0.5 a negative association with the outcome.The effect size was mainly used to compare the value, i.e., usefulness of measures that showed a significant association with the outcome.Spearman correlation was used to compute the correlation between time series.Square brackets are used to report interquartile ranges after median values.Data processing and analysis were performed using Python 3.8.

Results
Data from 30 patients with poor grade aSAH and extended neuromonitoring using the PbtO2 and NIRS probes (USZ: n=20; Charité Berlin: n=10) were analyzed.Thereof, 23 (USZ: n=16; Charité Berlin: n=7) had adequate data quality in all signals from PbtO2 and NIRS monitoring but for 4 of them data were not available in the time frames analyzed immediately before CT imaging.This resulted in a final number of 19 (USZ: n=14; Charité Berlin: n=5) patients.For the two time frames of 24 and 48 h before imaging, a total of 66 and 67 imaging events were analyzed, all of which stemmed from CT examinations.Due to the required minimum availability of 10 % of the data points, there is one fewer event that could be analyzed for the 24-hour time frame.Patient characteristics are given in Table 1.
The correlation analysis of the PbtO2 and rSO2 signals revealed only weak correlations between the signals (Fig. 1).The short-term correlation within the 15 min frames was 0.26 (IQR: [-0.15, 0.65]) and 0.16 (IQR: [-0.13, 0.56]) for left and right hemispheres, respectively.The  Of the 66 imaging events include in the analysis of the 24 h time frame, 7 events showed a new infarction and 47 events a perfusion deficit.Of the 67 image events of the 48 h time frame, 6 infarctions and 46 perfusion deficits where found.Where the difference stems from the inclusion criterion for the imaging events based on the data availability of the monitoring data of the two probe types.
Median PbtO2 measurements were significantly lower in patients who showed new infarctions in subsequent CT perfusion imaging for both time frames, with a better ROC-AUC value for the 24-hour time frame (0.812) compared to the 48-hour frame (0.751) (Table 2).The median of median PbtO2 values in the 24-h before new infarctions were detected was 17.3 [8,25] mmHg compared to 29 [22.5, 36] mmHg in patients with no new infarction.The NIRS parameters, on the other hand, showed no significant association with later detection of new infarctions.In contrast, median PbtO2 measurements were not significantly associated with a subsequent detection of a perfusion deficit for neither of the two time frames.However, in patients with subsequent confirmation of perfusion deficits, median HBoxy and HBtotal values Fig. 2. Median values for PbtO2 and NIRS measures for 24 and 48 h before imaging.Panels A-E of the first row depict boxplots for the median values of PbtO2, rSO2, HBoxy, HBdeoxy, respectively, computed over time intervals of 24 and 48 h before imaging.In these panels, boxplots are grouped by whether a new infarction was detected in the imaging or not (no (0), or yes (1)).Similarly, panels F-J show the boxplots for values grouped by whether a perfusion deficit was diagnosed from the imaging or not (no (0), or yes (1)).An asterisk (*) indicates a significant difference between the median values of the groups (p < 0.05).[62.4, 74.3] µmol/l, respectively.The differences in median values between the groups for the different parameters and the two time frames are also summarized as boxplots in Fig. 2. A high PbtO2 burden was significantly associated with new infarcts at all thresholds for the 24-hour frame, whereas the best differentiation was achieved with a threshold of 12 mmHg showing a ROC-AUC value of 0.203.Panel A in Fig. 3 shows the distributions for these burden values in the form of violin plots.For the 48-hour frame, ROC-AUC values improved with higher thresholds, but only the burden computed for the highest threshold of 32 mmHg was significantly associated with the occurrence of new infarction.Hypoxia burden calculated from PbtO2 was not significantly different between periods with and without subsequent perfusion deficits.Hypoxia burden computed for the NIRS parameters were not associated with new infarctions for all thresholds.However, certain burden values extracted from HBoxy, Hbtotal and rSO2 were significantly associated with perfusion deficits.For rSO2, the burden computed in the 48-hour frame was associated with thresholds above 76 % (Fig. 3B).For the 24-hour frame there was no association detected.Regarding HBtotal, the burden was significantly associated with all thresholds and both time frames.The optimal threshold was 74 µmol/l for showing ROC-AUC values of 0.286 and 0.279 for the 24-and 48-hour frames, respectively.HBoxy burden was significantly different between patients with and without perfusion deficit for thresholds starting from 44 µmol/l, and 46 µmol/l for the 24-and 48-hour frames, respectively.The burden values from HBoxy and HBtotal for the 48-hour frame are displayed in Fig. 3C and D, respectively.The best differentiation could be achieved using a threshold of 46 µmol/l (ROC-AUC: 0.321), and 58 µmol/l (ROC-AUC: 0.286), respectively.HBdeoxy burden showed no association with any outcome.The p-values from the burden analyses and all signals are displayed in Fig. 4.

Discussion
Our results confirm PbtO2 as a sensitive marker of cerebral infarction in comatose patients after severe aSAH.In the two time frames of 24 and 48 h before imaging diagnosed infarction, the recorded PbtO2 values were significantly lower.These results are consistent with the currently used threshold for PbtO2 in clinical practice which ranges from 20 to 25 mmHg (Sandsmark et al., 2012;Oddo et al., 2014).
However, with respect to the detection of cerebral hypoperfusion, the parameters HBoxy and HBtotal, determined by intraparenchymal NIRS measurement, proved to be more sensitive.
While we were able to demonstrate significant differences in these parameters with respect to imaging-detected perfusion deficits, the values for PbtO2 did not show any significant differences.
Several studies have explored the use of non-invasive NIRS in aSAH patients to detect changes in cerebral oxygenation for the prediction of ischemic events.However, in a recent review of the use of NIRS in patients with poor grade SAH, Francoeur et al. conclude that its use is currently not recommended in this population (Francoeur et al., 2022).
With non-invasive NIRS patches, the light penetrates the extracranial tissues such as skin, skull and dura which leads to inaccurate readings for the brain tissue (Davies et al., 2019;Sørensen et al., 2015) and prohibits the determination of absolute concentrations of the chromophores.Therefore, a brain tissue probe to perform NIRS directly in the brain tissue has been developed, which allows to eliminate extracerebral contamination (Keller et al., 2024).The accuracy of absolute values measured even with the probe can be compromised due to changing light scattering properties and deep path length factors in brain tissue.To overcome these limitations, extensive Monte Carlo simulations were performed as a basis for the development of the NIRS ICP probe (Mudra et al., 2024(Mudra et al., , 2006;;Boecklin et al., 2024).Although Hboxy and Hbtotal were associated with cerebral perfusion deficits, no significant differences for rSO2 in detection of perfusion deficits or new infarctions were found.An explanation for the poorer sensitivity of rSO2 compared to the Hboxy and Hbtotal might be a defective cerebral autoregulation in the majority of patients.Thus, in the area of impaired perfusion, Hboxy and Hbtotal might decrease in parallel and rSO2, as the coefficient between Hboxy and Hbtotal might remain unchanged.The fact that median Hbtotal values were significantly lower with perfusion deficits supports this hypothesis.Furthermore, despite our simultaneous and same location measurement of PbtO2 and rSO2, we could not show any clinically relevant correlation between these two.Recently, Hugues de Courson et al. ( 2022) also failed to demonstrate a correlation between intraparenchymal PbtO2 and transcranial measured rSO2 in a retrospective study of 51 SAH patients.The main hypothesis explaining the lack was again the influence of the extracranial tissue in the NIRS measurement.Since this influence is excluded in our measurement approach, we rather assume that a basic comparability is not possible due to measuring different aspects of cerebral oxygenation.While NIRS measures the concentration of oxygenated and deoxygenated hemoglobin in the brain tissue (mainly from capillaries, arteries and veins), PbtO2 reflects the partial pressure of dissolved oxygen in interstitial white matter.
Instead of rSO2 as a composite value, we focused on Hboxy as an absolute value reflecting the amount of oxygenated blood in the illuminated tissue.Changes in Hboxy, reflecting O2-delivery, are directly proportional to changes in CBF (Taussky et al., 2012).With a decrease in CBF, as with evidence of a perfusion deficit, there is a reduction in Hboxy.The significant differences in the levels of Hboxy 24 and 48 h before the occurrence of perfusion deficits in our patients strengthen this hypothesis.For Hbdeoxy, we found no significant changes in levels for either new infarctions or perfusion deficits.However, this then changed for Hbtotal.As with Hboxy, we found significant differences in the two studied time periods before detection of a perfusion deficit.Assuming that hemoglobin serves as a surrogate for CBV, an initial decrease is plausible, especially in severe vasospasm.
Intraparenchymal NIRS applied via brain tissue probes might be a promising technique for the detection of DCI in unconscious patients after severe aSAH.Focusing on Hboxy, reflecting CBF, and Hbtotal, reflecting CBV, rather than on rSO2 might allow to detect cerebral hypoperfusion early at critical thresholds of about 46 µmol/l.
A limitation of our study was that despite the two-center design, only a low number of patients was included.It remains technically very demanding to collect and temporally synchronize the high-resolution data in these complex patients.Therefore, data from only 19 of the original 30 patients could be analyzed.However, by using multiple events/CT images per patient, 66 or 67 events could be analyzed, increasing the power of the results.
As another limitation there was no existing protocol for the exact placement, timing of insertion and number of probes.Thus, the placement was uni-and bilaterally as well as on the affected and unaffected hemispheres.Overall, the new infarcts and perfusion deficits detected by imaging were not explicitly attributed to the probe location.This is directly related to the ubiquitous problem of multimodal neuromonitoring, which can only cover a small local region.Compared to the Licox probe, which covers a few cubic millimeters, the NIRS-ICP probe illuminates about 2-3 cubic centimeters of brain tissue (Boecklin et al., 2024).This provides a larger sample volume.In contrast, perfusion imaging with its coverage of most of the vascular supply areas is obviously superior.At the other hand, small focal ischemia, might be missed in perfusion CT.In our patients, imaging also revealed perfusion deficits Fig. 4. P-values of Mann-Whitney test for hypoxia burden.Panels A-E show the p-values corresponding to the associations between hypoxia burden and new infarctions, whereas Panels F-J show the p-values for the association between hypoxia burden and perfusion deficits.In each case, the burden was computed based on different parameters and depicted for a range of thresholds.The x axes are labeled in the units of measurement of the corresponding parameter.A p-value below 0.05 was interpreted as a significant association and indicated by a dashed line in the plot.that were not detected by local neuromonitoring.These patients were also labeled as DCI positive, so that the evaluation was altered if the neuromonitoring values were normal.As described, imaging was also performed when there was evidence of significant vasospasm due to increased flow velocities in the TCDs.Also, there exists a significant imbalance between the number of detected infarctions and perfusion deficits.However, the imbalance might be due to the fact, that we only classified truly new infarcts as such, while perfusion deficits might occur at any given day and location.

Conclusion
NIRS applying brain tissue probes might be a promising technique to detect cerebral perfusion deficits in patients with aSAH.Measurements in larger collectives are needed for validation.

Instructions to authors
We have reviewed the manuscript and verified that it complies with all the instructions provided to authors.

Authors approval
We confirm that all authorship requirements have been met and all individuals who contributed to the research have been appropriately credited as authors.We have obtained approval from all authors regarding the final version of the manuscript before submission.
Furthermore we confirm that this manuscript has not been published elsewhere and is not under consideration by another journal.

Fig. 1 .
Fig. 1.Correlation between PbtO2 and rSO2 measurements.Panels A and B show scatter plots of the 15 min medians for left and right hemispheres.
J.F.Willms et al.   correlation between the median values of the 15 min frames were 0.27 and 0.03 for left and right hemispheres, respectively.
were significantly lower for both time frames.The association was slightly stronger for the 48-hour time frame for which the median of median HBoxy and HBtotal were 52.9 [48.1, 55.1] µmol/l and 75.0 [70.0, 80.0] µmol/l compared to 45.9 [39.4,51.5] µmol/l and 69.5

Fig. 3 .
Fig. 3. Violin plots for selected hypoxia burden values for PbtO2 and NIRS measurements.Panels depict violin plots for the hypoxia burden values computed for the respective signals, time frames and different threshold values.The values are grouped by the corresponding outcomes, where red (blue) indicates the occurrence (absence) of a new infarction or perfusion deficit.The median values are indicated by a horizontal line.Panel A shows the burden computed from the PbtO2 for a time frame of 24 h.Panels B, C, and D depict the burden computed from the rSO2, HBoxy and HBtotal values, respectively, using the time frame of 48 h before imaging.

Table 1
Patient Characteristics.

Table 2
Overview of median values and the effect sizes measured by the ROC AUC computed for all parameters in the two time frames of 24 and 48 h before imaging.
Significant differences between median values are indicated in bold.Median PbtO2 values are stated in units of mmHg, HBoxy, HBdeoxy as well as HBtotal in units of µmol/l and rSO2 in percent.