Quantitative EEG for the detection of brain ischemia

Monitoring in most intensive care units (ICUs) is limited to continuous assessments of cardiopulmonary function, whereas brain monitoring has traditionally been limited to serial neurological examinations and infrequent imaging studies. Increasingly it is becoming clear that secondary neurological complications, such as seizures and brain ischemia, are also seen in the medical-surgical ICU population and are not limited to patients with primarily neurological injury. Electroencephalography (EEG) offers a continuous, real-time, non-invasive measure of brain function. Originally developed for the characterization of seizures and epilepsy, continuous EEG monitoring (cEEG) has been used for seizure detection in the ICU. Additionally, cEEG has been used as a method of identifying subclinical brain injury during neurosurgical procedures, such as carotid endarterectomy, and for ischemia detection, global function assessment, medication titration, and prognostication [1].


Introduction
Monitoring in most intensive care units (ICUs) is limited to continuous assessments of cardiopulmonary function, whereas brain monitoring has traditionally been limited to serial neurological examinations and infrequent imaging studies. Increasingly it is becoming clear that secondary neurological complications, such as seizures and brain ischemia, are also seen in the medical-surgical ICU population and are not limited to patients with primarily neurological injury. Electroencephalography (EEG) off ers a continuous, real-time, non-invasive measure of brain function. Originally developed for the characterization of seizures and epilepsy, continuous EEG monitoring (cEEG) has been used for seizure detection in the ICU. Additionally, cEEG has been used as a method of identifying subclinical brain injury during neurosurgical procedures, such as carotid endarterectomy, and for ischemia detection, global function assessment, medication titration, and prognostication [1].
In the ICU, ischemia detection in particular has been under-utilized, but the potential to diagnose ischemia as it occurs is tremendous (e.g., monitoring after high risk vascular or cardiac surgery, during refractory hypotension, or in the context of sepsis-associated encephalopathy). When cerebral blood fl ow (CBF) becomes compro mised, changes occur in both the metabolic and electrical activity of cortical neurons, with associated EEG changes [2]. In the operating room, EEG has an established role in identifying ischemia prior to the develop ment of infarction during carotid endarterectomy [3]. In acute ischemic stroke, the primary injury has typically occurred prior to presentation, but EEG may be able to detect patterns to suggest severity, prognosis, and secondary injury (e.g., reocclusion, edema, or hemorrhagic transformation) [4]. Delayed cerebral ischemia from vasospasm after subarachnoid hemorrhage (SAH) illustrates an application in which early detection may prevent the development of permanent damage by triggering appropriate interventions such as angioplasty or intra-arterial administration of vasodilator therapy [5,6]. Serial neurological exams and imaging are only capable of detecting delayed cerebral ischemia once the damage becomes clinically or radiographically apparent. In this case, EEG may be a useful way to detect and subsequently treat ischemia before the injury becomes irreversible [7][8][9][10].

EEG changes in ischemia
Brain function is represented on EEG by oscillations of certain frequencies. Slower frequencies (typically delta [0.  or theta [4][5][6][7]) are generated by the thalamus and by cells in layers II-VI of the cortex. Faster frequencies (or alpha, typically 8-12 Hz) derive from cells in layers IV and V of the cortex [11]. All frequencies are modulated by the reticular activating system, which corresponds to the observation of reactivity on the EEG [12]. Pyramidal neurons found in layers III, V, and VI are exquisitely sensitive to conditions of low oxygen, such as ischemia, thus leading to many of the abnormal changes in the patterns seen on EEG [4].
EEG changes are closely tied to CBF ( Figure 1) [3]. When normal CBF declines to approximately 25-35 ml/100 g/min, the EEG fi rst loses faster frequencies, then as the CBF decreases to approximately 17-18 ml/100 g/min, slower frequencies gradually increase. Th is represents a crucial ischemic threshold at which neurons begin to lose their transmembrane gradients, leading to cell death (infarction). In the setting of carotid clamping, CBF that decreases instantaneously to the ischemic threshold leads to rapid and reversible changes in the EEG (within 20 seconds) [3]. Infarction may not occur for hours at this degree of fl ow limitation [2] and some electrical activity (mostly delta frequencies) may be seen, but as the CBF continues to decrease toward the infarction threshold (10-12 ml/100 g/min and below), the EEG becomes silent and cellular damage becomes irreversible [2][3][4]. Some EEG patterns, such as regional attenuation of faster frequencies without delta (RAWOD), may refl ect early severe loss of CBF (mean 8.6 ml/100 g/ min) as seen during large occlusive infarcts, leading to complications such as edema and herniation [13].
Studies of CBF and cerebral rate of oxygen metabolism (CMRO 2 ) using Xenon computed tomography (CT) and positron emission tomography (PET) in ischemic stroke have demonstrated that regional EEG changes also refl ect the coupling of CBF and metabolism [14]. In early subacute ischemic stroke, the EEG correlates well with CBF as the oxygen extraction fraction increases to preserve CMRO 2 , a period termed "misery perfusion" or stage 2 hemodynamic failure [14,15]. Later, the EEG appears to correlate less well with CBF and instead begins to correlate with CMRO 2 during a period of "luxury perfusion" or stage 3 hemodynamic failure [14,15]. Early in this dynamic relationship, the threshold for cellular damage is shifted such that neuronal loss and decreased protein metabolism may in fact precede changes related to critical CBF, particularly in peri-infarct areas [2]. Th ese areas are also complicated by abnormal glutamate release (between 20-30 ml/100 g/min) or peri-infarct depolari zations [2,16].
Th e transition from ischemia to infarct occurs over a range of CBF, potentially providing a window of opportunity to treat and reverse impending neuronal cell death.
Diff usion-weighted magnetic resonance images (DWI MRI) are capable of detecting changes at CBF 35-40 ml/100 g/min within 30 minutes [17]. In contrast, EEG detects changes at the same CBF within seconds and allows for continuous monitoring of these changes over time. Th is can be crucial to detect evolving ischemic changes after treatment with tissue plasminogen activator (tPA), when the CT is negative during early infarction, or when there is a mismatch between DWI MRI and the clinical examination. Additionally, EEG may assist in guiding treatment, to rapidly reverse the region at risk if it has not reached the infarction threshold. Examples include shunt placement in carotid endarterec tomy [3], tPA in acute ischemic stroke [18,19], and aug mentation of blood pressure or CBF after acute ischemic stroke [20].
Th ere are some data suggesting that EEG may help with prognostication in patients with acute ischemic stroke. Because cortical changes are modulated by brainstem reticular formations, global changes, such as loss of reactivity [21] and lack of sleep-wake cycles [22], may portend poor prognosis (and indicate possible brainstem involvement). Regional changes, including a lack of delta or the presence of faster frequencies within 24 hours, correlate with a good outcome [23] while the persistence of unilateral prominent, continuous delta slowing or decreased alpha adds signifi cantly to clinical predictions of poor outcome [24]. Other specifi c patterns may also indicate poor prognosis, such as RAWOD (with up to 67% mortality) [13] or periodic lateralized epileptiform discharges (PLEDs) [25].

A need for quantitative EEG
Despite the potential to detect ischemia, there are a number of limitations that need to be overcome before using this technique for automated event detection in ICU practice. Early obstacles, such as data acquisition, remote access, and data storage have been largely overcome by technological advances. Nonetheless, raw EEG requires interpretation by trained experts, ideally with experience in continuous monitoring for critically ill patients. Subtle changes over time (hours to days) may be missed and visual analysis can be extraordinarily timeconsuming. To achieve the goal of real-time detection of critical brain events, raw EEG would require continuous expert review. Although telemedicine has become a prominent tool in acute stroke, centralized EEG monitor ing with access to around-the-clock neurophysiology experts is not currently available and would be extremely expensive.
Th e most pressing limitation to cEEG remains its interpretation, which is ultimately subjective. By applying a Fourier transformation, EEG can be quantifi ed in terms of its amplitude, power, frequency, and rhythmicity in order to generate numerical values, ratios, or percentages; graphically display arrays or trends; and set thresholds for alarms. A variety of quantitative EEG (qEEG) measures have been used clinically to quantify slowing or attenuation of faster frequencies in the EEG, specifi cally: Th e calculation of power within diff erent frequency bands (i.e., delta, theta, alpha, and beta); ratios or percentages of power in specifi c frequency bands; and spectral edge frequencies (based on the frequency under which x% of the EEG resides). Th ese discrete values can then be compared between diff erent regions (between hemispheres, for instance, or between electrode-pair channels). Time-compressed spectral arrays were developed to incorporate both power and frequency spectrum data and are reconstructed using color to represent power at diff erent frequencies (a so-called 'spectrogram'). Th ese spectrograms have the highest pixel-to-data ratio, yielding easy-to-digest color representations of large amounts of data within a single screen (compared with the raw EEG) [26]. Additional measures include amplitude integra ted EEG, which was developed to continuously monitor comatose patients via average ranges of peak-topeak amplitudes displayed using a logarithmic scale, and the commercial Bispectral Index, often used to monitor the depth of anesthesia (Aspect Medical Systems, Inc., Newton, MA). Other nonparametric methods exist beyond Fourier transformation, including interval or period analysis and alternative transformation techniques. Parametric, mimetic, and spatiotemporal analyses are also available using a variety of computational methods that are beyond the scope of this review [27]. For now, fast Fourier transformation appears to be the most suitable for quantifying the EEG [28] but future algorithms may also incorporate ICU-specifi c EEG feature recognition and waveform analysis based on machine learning approaches trained on ICU EEG recordings.
It is important to emphasize that there are a large number of confounders to be considered when evaluating EEG signals. Both raw EEG and qEEG can be altered in response to the presence of seizures or periodic discharges, changes in intracranial pressure (ICP, e.g., acute hydrocephalus), or in the setting of systemic illness such as ventilator-acquired pneumonia (VAP). In addition, sedatives, narcotics, antipsychotics, and other medications may have varying eff ects on the EEG. Even normal state changes, such as slow wave sleep, appear similar to changes seen during ischemia. Moreover, the ICU is contaminated with a variety of artifacts: Chest percussion, bed movement, condensation in ventilator tubing, and electrical (50Hz or 60Hz) artifact from other monitoring devices. Ultimately, the raw EEG remains the gold standard for the assessment of brain function. Any algorithm based on qEEG should include raw EEG for review by staff familiar with EEG -either on site or remotely [9,26].

Quantitative EEG in ischemia
Like raw EEG, qEEG is capable of refl ecting changes in blood fl ow and metabolism in as little as 28-104 seconds [29]. However, diff erent quantitative parameters may have their own strengths for a given clinical situation. Intra operative spectral edge frequencies, for example, detect ischemic changes in carotid endarterectomy patients receiving isofl urane anesthesia, whereas changes in the relative delta percentage (that is, delta power/total power) are accurate in patients receiving propofol [29]. In acute ischemic stroke, hemispheric relative delta percentage, spectral edge frequencies 25% and 75%, and the overall mean frequency correlated well with CBF [28]. Cerebral perfusion pressure (CPP), related to CBF and to ICP, correlates with decreases in mean frequency as CPP falls below 60 mmHg, even in comatose or sedated patients [30]. Finally, increased power in slower frequency bands (delta and theta) and decreased power in faster frequency bands (alpha and beta) are seen with reduc tions in brain metabolism (CMRO 2 ) [14]. Overall, the rela tive delta percentage appears to provide the most robust correlation with CBF and metabolism during focal ischemia.
Clinically, qEEG correlates with stroke severity, radiographic fi ndings, and response to treatment. Several parameters correlate with initial stroke severity as measured by the National Institutes of Health Stroke Scale (NIHSS) in both the acute (brain symmetry index [BSI]) [19,31] and subacute periods (relative alpha percen tage, relative alpha-beta percentage, relative delta-theta percentage, delta/alpha ratio, delta-theta/alpha-beta ratio, and global pairwise derived BSI [pdBSI]) [32][33][34]. qEEG parameters correlate with the volume of infarc tion on MRI, such as the acute delta change index [35], global pdBSI [32], relative alpha percentage, relative alpha-beta percentage, relative delta-theta percentage, delta/alpha ratio, and delta-theta/alpha-beta ratio [36]. In the case of subcorti cal infarcts, both the delta-theta/alpha-beta ratio and the pdBSI correlate surprisingly well with infarct volumes, possibly related to the consequences of diaschisis on CBF in lacunar brainstem infarcts [33,36]. qEEG may be more sensitive than raw EEG to subtle changes [37,38], and some parameters may even detect improvement prior to improvement in the clinical exam. For example, after tPA, a decrease in delta power was documented 100 minutes prior to the start of functional improvement [18] and in a series of 16 patients with moderate stroke, one patient had an improvement in BSI a few minutes prior to clinical improvement after the administration of tPA [19].
qEEG may also provide a method of quantifying shortterm and long-term prognosis. At one year, functional outcome after stroke in one series was predicted correctly by clinical criteria 60% of the time. Th e addition of EEG data improved the predictive value to 85% (for both raw and quantitative EEG) [24]. Th e acute delta change index correlates with 30-day NIHSS as accurately as initial mean transit time on MRI and actually better than initial DWI volume (emphasizing the link between qEEG parameters and blood fl ow) [35]; other parameters such as the delta/alpha ratio and the relative alpha percentage also correlate with the 30-day NIHSS [34]. As early as 6 hours after stroke, the pdBSI correlates with the 1-week NIHSS. Subacute abnormalities in pdBSI and, particularly, the delta-theta/alpha-beta ratio are associated with the 6-month modifi ed Rankin score [39]. For deeper infarcts, pdBSI and the delta-theta/alpha-beta ratio correspond to 1-week outcome [33]. Extremely elevated pdBSI may even predict early neurodeterioration from other complications at 1 week [36].
Th e ability to infer CBF patterns, quantify the severity of brain ischemia, and objectively monitor the response to treatment may be useful across critically ill patient populations. Ischemia detection could alert intensivists to acute ischemia after cardiac valve replacement, for example, and could prompt intra-arterial thrombectomy and prevention of stroke. Titration of vasopressors may be infl uenced by the detection of ischemic changes despite an otherwise appropriate mean arterial pressure (MAP) goal. Although recent studies of EEG in patients with sepsis have focused on seizure detection and other abnormal epileptiform patterns [40], quantifi cation of cerebral perfusion in these patients could be an invaluable measure of end-organ perfusion to guide pressor management, particularly in patients with sepsis-associated encephalopathy, which may be related to impaired perfusion of the small vessels [41]. SAH provides a useful example to illustrate the need for and potential of existing data related to qEEG techniques in the ICU.

Clinical scenario: delayed cerebral ischemia after subarachnoid hemorrhage
SAH has an incidence of 6-15/100,000 and is associated with high morbidity and mortality [42]. Much of the outcome is determined by secondary complications many of which are treatable if diagnosed in a timely fashion. Between 20-40% of SAH patients either clinically deteriorate or have evidence of new ischemia on repeat brain imaging from vasospasm, together termed delayed cerebral ischemia [5,42,43]. Delayed cerebral ischemia is associa ted with worse outcome [5], but may be particularly challenging to diagnose in the comatose patient [6].
Th e eff ects of vasospasm-mediated delayed cerebral ischemia can be mitigated with pressure and volume augmentation (so-called triple-H therapy), intra-arterial vasodilators, such as verapamil, or angioplasty. Th e bene fi t of these measures relies on the earliest possible detectionbefore neuronal damage becomes irrever sible. Serial neurological exams and radiological confi r ma tion of ischemia off er post hoc diagnosis, but do not adequately detect preclinical events. Currently, the only serial detection method in routine use is transcranial Doppler ultrasonography (TCD). Although mean blood fl ow velocities of > 120 cm/s are around 60-70% sensitive to changes in the proximal middle cerebral artery (MCA), this cut-off is less reliable (and less well studied) for other territories and does not include the distal vasculature [42]. When used daily, TCDs have been 60-70% sensitive for delayed cerebral ischemia with limited predictive value; they also exhibit frequent false negatives or false positives [44].

EEG changes in subarachnoid hemorrhage
SAH produces a wide variety of abnormalities on EEG. Even before the onset of vasospasm, EEG may detect disorganization or slowing, disruption of the normal posterior dominant rhythm, seizures (6%), periodic discharges (16%), a lack of reactivity (14%), or a lack of sleep transients (in up to 85%) [45,46]. In a serial EEG study of 151 patients with SAH, the authors found that even on day 1, certain patterns predicted later vasospasm, including the presence of very brief frontally predominant biphasic delta waves ('axial bursts'), focal polymorphic delta overlying an area of clot, or a predominance of unreactive delta frequency slowing. By day 5, the development of continuous polymorphic, rhythmic, or unreactive delta predicted vasospasm 100% of the time. Th ese patterns correlated well with CT grades of SAH severity and 40% of patients with abnormal EEG on day 5 had neurodeterioration [47]. As predicted by the EEG's close correlation with CBF and metabolism, raw EEG was capable of detecting developing ischemia up to 78% of the time in one study of poor-grade SAH patients [9].
Th e use of qEEG for ischemia detection in SAH has been examined systematically in a total of 89 patients  (Table 1). In the fi rst of these studies, EEG was found to detect ischemia earlier and better than clinical exam alone [7] in 11 patients, most of whom were Hunt-Hess grade 3. Total power, the alpha/ delta-theta ratio, and the relative delta percentage were performed in only six patients, but were in aggregate 100% sensitive to detect infarcts if the changes were sustained over a 24-hour period. Total power specifi cally detected 100% of the silent infarcts seen on CT; compressed spectral array on the other hand, was 67% sensitive for ischemia. Four ischemic events had qEEG changes prior to the development of clinical defi cits and fi ve ischemic events had changes in total power that predicted silent infarcts when the clinical exam did not. Th ere were only two false positives reported during 62 days of monitoring.
qEEG and xenon CT measures of CBF were used simultaneously in 32 awake SAH patients (Hunt-Hess 1-3) to detect vasospasm, which was confi rmed by angiography or TCD [8]. Th e relative alpha percentage, defi ned uniquely as 6-14 Hz frequency power divided by the total power (1-20 Hz), was calculated over 2 minute windows to generate a histogram periodically throughout the day. Th e variability of this histogram was then qualitatively scored from 1 to 4. Decreased variability trended toward decreased CBF. Overall, 19 of 32 patients developed vasospasm, all of whom had decreased variability of the relative alpha percentage. Seven patients without vasospasm had decreased variability, largely due to stroke or increased ICP. Changes in variability occurred in 4 or more of the 6 electrode channels monitored, suggesting this measure refl ects a more global change. Importantly, decreased variability occurred a mean of 2.9 days prior to the development of TCD or angiographic vasospasm.
Using a standard electrode array divided into vascular distributions (anterior circulation and posterior circulation), 34 comatose SAH patients (Hunt-Hess 4-5) were evaluated to detect delayed cerebral ischemia; angiography and TCD data were not required to confi rm the clinical or radiological diagnosis [9]. Rather than utilizing continuously generated data, 10 sixty-second clips were obtained both at baseline and between days 4 and 6. Th e strength of this particular analysis is that each clip was obtained after stimulation -thereby eliminating the variability associated with state changes or sleep. Total power and power in each frequency band were measured in addition to relative alpha, relative delta, the alpha/delta ratio, and the alpha-beta/delta ratio. Th e coherence of each hemisphere was calculated for both alpha and delta frequencies and a total average frequency was also determined. Diff erent cut-off s were used to determine whether a change was signifi cant: 10, 20, or 50% changes from baseline. In predicting the 9 of 34 patients with delayed cerebral ischemia, the post-stimulation alpha/ delta ratio correlated best when anterior and posterior regions were considered together, followed by relative delta, alpha-beta/delta ratio, and relative alpha (Figure 2). For region specifi c parameters, only the relative delta percentage and the alpha/delta ratio correlated with delayed cerebral ischemia and only in the anterior derivations; however, only one patient had a posterior circulation event. When six minutes of EEG were used, a 10% change in the alpha/delta ratio was 100% sensitive and 76% specifi c for delayed cerebral ischemia. When just one minute was used, a 50% or more change from baseline yielded a sensitivity of 89% and specifi city of 84%. Overall, relative rather than absolute parameters appeared to be more sensitive to changes. Despite indications that hemispheric indices are predictive of clinical cEEG monitoring was performed from SAH days 3 to 8. The ADR progressively decreased after day 6, particularly in the right anterior region (blue arrow), to settle into a steady trough level later that night, refl ecting loss of fast frequencies and increased slowing over the right hemisphere in the raw cEEG (c; EEG2 compared with EEG1). On SAH day 6, fl ow velocities in the right MCA were marginally elevated (144 cm/s), but the patient remained clinically stable with hypertensive, hypervolemic therapy (systolic blood pressure >180 mmHg). On day 7, the GCS dropped from 14 to 12 and a CT scan showed a right internal capsule and hypothalamic infarction (b, Day 7). Angiography demonstrated severe distal right MCA and left vertebral artery spasm; however, due to the marked tortuosity of the parent vessels and the location of vasospasm, a decision was made not to perform angioplasty, but to infuse verapamil and papaverine. This resulted in a marked, but transient increase of the right anterior and posterior alpha/delta ratios (blue shaded area). Later that day the patient further deteriorated clinically to a GCS of 7, with a new onset left hemiparesis, and died on SAH day 9 from widespread infarction due to vasospasm. From [9] with permission. Similarly, although spectral edge frequencies correlate with ischemia in acute ischemic stroke [28] and carotid endarterectomy [29] and mean frequency changes correlate with improvements in CBF [20], frequency measures were not predictive of delayed cerebral ischemia in poorgrade SAH, potentially because of lower overall frequencies at baseline. Th ese parameters have not been adequately studied in better grade SAH, and may still hold some promise. A recent study used a standard electrode array, excluding the posterior derivations to avoid the impact of the frequent (albeit normal) asymmetry of the posterior dominant rhythm on qEEG values [10]. Twelve patients with modifi ed Fisher grades 3-4, corresponding to highest risk for vasospasm [43], were evaluated using the total power of the alpha frequency band (defi ned as 8-15 Hz). Each 30 seconds, an alpha frequency value was calculated and averaged over 30 minutes. Th e standard deviation and the mean alpha power were then multiplied to create a 'composite alpha index' . Overall, 8 of 12 patients developed delayed cerebral ischemia. Th ree patients with delayed cerebral ischemia developed worsening trends in the composite alpha index more than 24 hours prior to neurodeterioration. Predictions of neurodeterioration made exclusively on clinical data were 40% sensitive, whereas the addition of qEEG data increased this to 67% with a specifi city of 73%.

Future directions
Both the relative alpha variability and the composite alpha indices were derived in an eff ort to generate a discrete trending value accessible to other members of the patient care team. Unfortunately, both measures require artifact detection based on visual inspection by electroencephalographers. While post-stimulation clips can be easily evaluated for artifact at the time of the stimulation, this may be labor intensive and does not provide a continuous measure. Future studies of ischemia detection should focus on discrete and easily interpretable values generated in real-time, accessible to the neurophysiologist, intensivists, and the nursing staff .
Based on qEEG data in ischemia, it appears that regional power values may be more specifi c to the nature and size of the underlying lesion and, therefore, correlate with later functional outcome; however, they may not be sensitive enough to use alone in ischemia detection [9]. Hemispheric power diff erences highlight acute ischemic changes, but correlate less well with the subtle, multifocal, and more gradual changes seen in SAH. qEEG frequency values (such as spectral edge frequencies) may be useful to detect changes in relatively healthy brain [29], but it is not clear that these are helpful in more widespread pathology such as SAH. Some values (such as global increases in theta range frequency) may have a relationship to the penumbra and, therefore, be useful in the pre-infarct stage of ischemia in good-grade SAH [36] but this has not been adequately studied. To develop a rational approach to the detection of ischemia, a combination of regional, hemispheric, and total measures of both power and frequency may be needed.
An attempt should be made to integrate qEEG parameters into the multimodal monitoring infrastructure available in some ICUs. Electrophysiological changes should be interpreted in the context of blood pressure, systemic oxygenation, intracranial or cerebral perfusion pressures, cerebral measures of oxygen tension, or microdialysis. More recently, intra-cortical electroencephalography has provided valuable information on both seizure and ischemia detection: In a series of 5 patients who had intra-cortical electroencephalography, a change in the alpha/delta ratio of 25% sustained over 4 hours was the most accurate parameter in detecting vasospasm between 1 and 3 days prior to confi rmation with angiography [48]. Subdural electrodes have detected periinfarct depolarizations and cortical spreading depression [16], phenomena invisible to scalp EEG but that might play a role in delayed cerebral ischemia after SAH. Interestingly, using subdural electrodes these researchers were also able to detect an abnormal hemodynamic response when analyzing the cortical spreading depression signal in relation to ischemia from vasospasm [16].
Ultimately, ischemia detection requires real-world validation. Software currently exists that allows real-time quantifi cation of qEEG parameters. Th e numerical values generated by qEEG can be used to set threshold alarms such that when a specifi ed value is reached for more than a specifi ed period of time (e.g., to avoid confusion with artifact), an alert is sent to a treating member of the ICU team. Th is should be done in the context of a prospective and randomized trial that compares integrating qEEG into a novel algorithm versus best clinical practice, for example comparing diff erent triggers to obtain established confi rmatory assessments of vasospasm-associated delayed cerebral ischemia, such as CT angiography, CT perfusion, or conventional angiography.

Conclusion
EEG is a very promising tool for monitoring brain function in real-time in the ICU. Th ere are characteristic changes that occur on EEG in response to brain ischemia, correlating with CBF and brain metabolism. Although raw EEG evaluations are time consuming, quantifi cation of EEG features in real-time has been incorporated into many of the standard EEG software packages. Some of these quantitative parameters have also been found to correspond to physiologic changes associated with ischemia. Sensitive techniques are needed to detect cerebral ischemia, for example, in vasospasm-associated delayed cerebral ischemia after SAH. Early evidence suggests that qEEG may be sensitive enough to allow pre-clinical detection of delayed cerebral ischemia from vasospasm. Th is approach may be utilized to widen the window of opportunity in order to prevent permanent neuronal damage in a variety of clinic scenarios.