Intracranial electroencephalographic approaches in the intensive care unit: Safety, feasibility, and coverage

Abstract

Objective

Nonconvulsive epileptic activity is common after acute brain injury and contributes to neuronal injury and poor outcomes. Although intracranial electroencephalography (iEEG) improves detection compared with surface EEG (suEEG), it currently relies on focal recordings of epileptic dynamics. We prospectively evaluated multielectrode iEEG strategies designed to achieve large‐scale cortical and deep brain coverage in critically ill patients.

Methods

In a prospective cohort of adults with acute brain injury requiring invasive neuromonitoring, we implemented four depth electrode‐based iEEG techniques: three electrocorticographic approaches (open subdural, burr hole subdural, and foramen ovale) and one stereotactic (transcortical) approach. The primary objective was to establish and quantify the extent and durability of cortical recordings. Additional assessments included safety, feasibility, and management impact. Electrode localization was performed in MNI305 space, and recording longevity was assessed using Kaplan–Meier and Cox proportional hazards models.

Results

Thirty patients with severe brain injury underwent implantation of 64 electrodes (772 contacts), yielding 361 total iEEG recording days across heterogeneous etiologies and surgical settings. Multielectrode implantation enabled distributed sampling spanning multilobar frontoparietotemporal cortical clusters. Overall, 72% of electrodes sampled cortical convexity alone, whereas 28% incorporated deep targets, including mesial temporal and orbitofrontal regions. Recording durability differed by implantation strategy (log‐rank p = .0002), with a mean combined iEEG–suEEG monitoring duration of 14.5 days per patient. Implantation was successful in >95% of electrodes, monitoring started within 24 h postoperatively in 88%, and the per‐electrode risk of a causally related adverse event was 3.1%, without hemorrhagic complications. iEEG influenced treatment initiation or escalation in 32% of monitored patients.

Significance

Multielectrode iEEG strategies enable durable, large‐scale cortical sampling in acute brain injury while remaining feasible and safe in the intensive care unit. By shifting from focal to distributed spatial sampling, these approaches provide a conceptual framework for improving detection of epileptic activity in critically ill patients.

Key points.

• Multielectrode intracranial EEG strategies enable distributed sampling in critically ill patients with acute brain injury.

• Large‐scale cortical coverage can be achieved across frontal, parietal, and temporal regions, with optional sampling of deep structures.

• Cortical convexity is sampled with subdural depth electrodes, whereas deep targets are accessed via transcortical or foramen ovale approaches.

• These strategies are feasible in diverse neurosurgical settings, with high implantation success and excellent safety profiles.

• Recording durability is robust across techniques, although contact survival varies by implantation strategy.

1. INTRODUCTION

Multimodal neuromonitoring after severe brain injury is central to guiding intensive care management in comatose patients. Epileptic activity, ranging from electrographic seizures and status epilepticus to rhythmic and periodic patterns (RPPs), is common after acute brain injury and relies almost exclusively on electroencephalographic (EEG) detection, as these events are predominantly nonconvulsive and lack overt clinical manifestations. ¹ Using continuous surface EEG (suEEG), seizures are detected in approximately 10%–30% of patients and RPPs in 20%–40% of those with aneurysmal subarachnoid hemorrhage (aSAH), traumatic brain injury (TBI), intracerebral hemorrhage (ICH), or central nervous system infection (CNSi). ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ Importantly, these epileptic phenomena are associated with secondary brain injury and worse neurological outcomes in a dose‐ and time‐dependent manner. ³ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶

Despite its widespread use, continuous suEEG has inherent limitations in the critically ill population. Signal quality is frequently degraded by artifacts, requires daily maintenance, and is constrained by signal attenuation through the skull and scalp. As a result, suEEG preferentially detects epileptic activity only when discharges are synchronized across large cortical territories, whereas focal or deep epileptic activity, such as seizures arising from the hippocampus, may remain undetected. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ These limitations are particularly relevant in acute brain injury, where lesion topology and epileptic activity can be spatially restricted or focal.

To overcome these limitations, intracranial EEG (iEEG) has recently been introduced into the intensive care unit (ICU) as part of multimodal invasive neuromonitoring. Several studies have demonstrated that iEEG substantially increases the detection of seizures and RPPs compared with suEEG in patients with acute brain injury. ³ , ⁸ , ¹² , ²⁵ , ²⁶ , ²⁷ However, current ICU applications of iEEG typically rely on focal recordings, most commonly a single subdural strip electrode placed over perilesional cortex during craniotomy or a short intraparenchymal depth electrode inserted into frontal cortex. ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ Although these approaches are feasible and relatively safe, their spatial extent is limited.

At present, standardized implantation strategies aimed at expanding iEEG coverage while remaining feasible in urgent settings are lacking. Approaches that balance invasiveness with broader cortical and deep brain sampling have not been evaluated in challenging acute brain injury populations. In this context, we evaluated four depth electrode‐based iEEG strategies in acute brain injury: one stereotactic transcortical approach and three electrocorticographic (ECoG) techniques (open subdural depth electrodes, burr hole subdural depth electrodes, and foramen ovale electrodes). These approaches were designed to (1) extend spatial sampling across convexity and deep cortical regions, (2) provide applicability across a wide range of neurosurgical and ICU scenarios, and (3) deliver durable recordings while preserving safety and feasibility. Here, we prospectively establish the spatial extent and durability, feasibility and safety, and clinical impact of novel iEEG approaches. This provides an adoptable conceptual framework for invasive EEG in acute brain injury, which will allow the study of acute epileptic dynamics.

2. MATERIALS AND METHODS

2.1. Study design and population

We prospectively enrolled consecutive adult patients (≥18 years) with severe acute brain injury and impaired consciousness who required invasive neuromonitoring as part of standard care at a single academic center (University Hospitals Leuven, Belgium) between January 2024 and December 2025. Injury etiologies included aSAH, severe TBI (sTBI), ICH, and CNSi. This study is part of a prospective cohort (CONSCIUS study; ClinicalTrials.gov NCT06426615). Written informed consent was obtained from the legal representative, and ethical approval was granted by the local ethics committee.

The primary objective of this report was to quantify spatial sampling and ICU recording durability achieved by multielectrode iEEG strategies, as these are prerequisites for valid evaluation of diagnostic yield. Additionally, we assessed (1) safety (procedure‐related adverse events per patient and electrode), (2) feasibility (implantation and explantation success rates, time to monitoring, additional operative time), and (3) impact of iEEG on clinical decisions. Analyses of epileptic activity detection rates and comparison with suEEG are planned separately.

2.2. Implantation strategy and decision‐making framework

iEEG electrodes were implanted in the hyperacute setting during either (1) craniotomy or craniectomy procedures or (2) monitoring‐only procedures (burr hole‐ or bolt‐based invasive monitoring). Importantly, no patient underwent craniotomy or burr hole placement for the primary purpose of iEEG recording. Electrode implantation was only performed in patients already requiring neurosurgical intervention with invasive monitoring for clinical indications. Implantation of intracranial electrodes was combined with continuous suEEG (10–20 system). In this context, the incremental risk of adding intracranial electrodes to an already indicated invasive procedure was weighed against the high baseline risk of nonconvulsive epileptic activity and secondary brain injury in this population.

We implemented four implantation strategies that could be combined depending on clinical context: three ECoG (open subdural, burr hole subdural, foramen ovale) and one stereotactic (transcortical) approach. The selection of techniques followed a predefined decision‐making framework aimed at optimizing spatial sampling while minimizing invasiveness. Key considerations included (1) surgical setting (craniectomy vs. monitoring‐only), (2) injury localization and laterality, (3) presence of deep or temporal lesions, (4) anticipated need for secondary decompression, and (5) surgical convenience in the case of foramen ovale and transcortical electrodes.

Typically, approximately two electrodes were implanted per patient, preferentially in the most affected hemisphere, targeting both perilesional and nonlesional cortex while avoiding nonviable tissue. Subdural approaches were used to sample cortical convexity across multiple gyri and lobes, whereas foramen ovale and transcortical approaches were employed when deep or mesiotemporal regions were considered potentially relevant based on the presence of focal lesions. All implantations were performed by two neurosurgeons experienced in epilepsy surgery electrode techniques (S.S. and T.T.). A more detailed description of surgical decision‐making is provided in Methods S1.

2.3. EEG acquisition and monitoring

Following postoperative computed tomographic (CT) imaging, patients underwent continuous combined iEEG–suEEG monitoring using Micromed EEG systems (Natus). Monitoring continued until recovery of command following, death, or a maximum of approximately 3–4 weeks. Both iEEG and suEEG were used for clinical decision‐making and treatment of epileptic patterns, according to standard ICU protocols.

2.4. Electrode localization and spatial coverage

Electrode positions were localized using postoperative CT registered to MNI305 space using Advanced Normalization Tools (version 2.6.3) and 3D slicer (version 5.8.1). Cortical surface projections and anatomical labels were derived using the FreeSurfer fsaverage surface and Destrieux atlas (aparc.a2009s), respectively. ³⁰ Spatial coverage was summarized using cortical density maps, continuous kernel density estimates, and three‐dimensional visualizations. All figures were generated in Python (Python Software Foundation v3.12). A detailed description of the electrode localization protocol is provided in Methods S2.

To facilitate interpretation, we used a categorical pipeline (pathology—surgical setting—implantation technique—superficial and deep cortical regions—cortical clusters) where electrode contacts were first aggregated into simplified cortical regions reflecting predominant sampling patterns: frontal, parietal, and temporal superficial and deep cortical regions. Superficial and deep cortical regions were then combined into final cortical cluster labels based on multilobar coverage. In this context, the term cluster is used descriptively to identify recurring patterns of multilobar anatomical coverage. A schematic overview of implantation pathways and resultant coverage is provided in Figure S2.

2.5. Recording durability, safety, and feasibility

Recording durability was defined as time to contact failure. Kaplan–Meier curves were compared across implantation strategies using the log‐rank (Mantel–Cox) test. The proportional hazards assumption was assessed using Schoenfeld residuals. Cox proportional hazards models were fitted with robust standard errors clustered by patient to account for within‐patient correlation across contacts. Open subdural electrodes served as the reference category. Contact failure was assessed daily and defined as sustained loss of interpretable EEG signal with impedance > 1000 kΩ for ≥24 h. Analyses were performed in Python using the lifelines package. ³¹

Adverse events were prospectively recorded, categorized by causality, and summarized as per‐patient and per‐electrode risk. Feasibility metrics included implantation success rates (intention‐to‐implant principle), bedside explantation without reintervention, time to monitoring (between surgery and start of combined iEEG–suEEG monitoring), and additional operative time.

2.6. Electrophysiological findings and clinical impact

Routine daily clinical EEG reviews and clinical documentation as part of standard clinical care were used to descriptively classify epileptiform activity and to determine whether iEEG or suEEG findings were cited as drivers of antiseizure medication initiation or escalation. ³²

3. RESULTS

3.1. Cohort and procedural characteristics

Thirty patients with severe acute brain injury underwent electrode implantation (Figures 1 and 2). Etiologies included sTBI (70%), aSAH (17%), ICH (7%), and CNSi (7%; Figure 3A). A total of 64 intracranial electrodes were implanted (median = 2 electrodes per subject), yielding 772 recording contacts (mean = 26 intracranial contacts per subject). Thirteen patients (43%) underwent implantation during craniotomy or craniectomy, whereas seventeen (57%) underwent a monitoring‐only procedure. Four implantation strategies were used: open subdural (24 electrodes, 38%), burr hole subdural (22 electrodes, 34%), transcortical (14 electrodes, 22%), and foramen ovale (four electrodes, 6%; Figures 2 and 3B). Techniques were frequently combined within patients to extend spatial coverage and cover distributed perilesional areas (Figure 1).

FIGURE 1.

Implantation flowchart illustrating the decision‐making process for intracranial electroencephalography (EEG) in patients with severe acute brain injury (n = 30). The algorithm integrates (1) surgical setting (craniotomy with monitoring vs. monitoring‐only), lesion characteristics (convexity vs. deep/mesial involvement) and (2) feasibility constraints (presence of an external ventricular drain [EVD] or standalone intracranial pressure [ICP] bolt) to guide selection of implantation strategy. Patients undergoing craniotomy (left branch) were primarily implanted with open subdural depth electrodes, with optional foramen ovale (FO) electrodes when temporal sampling was indicated. In monitoring‐only settings (right branch), burr hole‐based approaches were favored, including burr hole subdural depth electrodes alone or combined with transcortical depth electrodes when deep targets were suspected. In selected cases without EVD, a transcortical approach was used. Implantation strategies were further stratified by (3) intended sampling depth (superficial convexity vs. combined superficial and deep); (4) laterality (unilateral vs. bilateral), where the most damaged hemisphere was envisioned; and (5) underlying pathology or lesion distribution, where the primary injured lobe was targeted. This framework enabled patient‐ and pathology‐specific optimization of spatial sampling while preserving feasibility in acute and heterogeneous surgical settings. Technical implantation details are provided in Methods S1. aSAH, aneurysmal subarachnoid hemorrhage; bilat, bilateral; CNSi, central nervous system infection; DAI, diffuse axonal injury; ICH, intracerebral hemorrhage; ipsilat, ipsilateral; IVH, intraventricular hemorrhage; TBI, traumatic brain injury; temp, temporal; unilat, unilateral.

FIGURE 2.

Overview of intracranial electroencephalographic (EEG) implantation strategies. (A) Schematic representations, postoperative computed tomographic (CT) images, intraoperative photographs, and representative bipolar EEG traces for each implantation strategy. For each example, paired surface EEG (suEEG) and intracranial EEG (iEEG) recordings are shown. suEEG traces are displayed in a standard bipolar montage (Fp2–F8, F8–T8, T8–P8, P8–O2). iEEG traces are shown in bipolar montage of adjacent contacts. Open subdural electrodes captured focal frontal electrographic seizures without a scalp correlate. Burr hole subdural electrodes demonstrated burst suppression visible on both iEEG and suEEG. Foramen ovale electrodes recorded mesiotemporal sporadic epileptiform discharges with limited frontotemporal scalp representation. Transcortical depth electrodes detected periodic discharges confined to deep temporal regions without scalp correlates. (B) Schematic representation of conventional focal iEEG approaches using a single subdural strip or short intracortical depth electrode.

FIGURE 3.

Cohort characteristics and spatial sampling. (A) Etiology of acute brain injury among included subjects (n = 30). (B) Distribution of implantation techniques across all implanted electrodes (n = 64). (C) Superficial cortical targets sampled per subject, demonstrating broad convexity coverage across frontal, parietal, temporal, and occipital regions. (D) Deep cortical targets sampled in a subset of subjects (n = 9), including mesial temporal, frontal, parietal, and insular regions. (E) Subject‐level coverage of cortical clusters, categorized according to the number of major lobes sampled per patient (frontal‐only, frontoparietal, frontotemporal, or frontoparietotemporal). (F) Within‐cluster sampling density, expressed as the percentage of electrode contacts located in frontal, parietal, and temporal regions for each cluster configuration. Together, these panels illustrate that multielectrode implantation strategies enable distributed, multilobar cortical sampling, including selective deep targets, across a heterogeneous acute brain injury population. aSAH, aneurysmal subarachnoid hemorrhage; CNSi, central nervous system infection; ICH, intracerebral hemorrhage; sTBI, severe traumatic brain injury.

Implantation was unilateral in 80% of patients and bilateral in 20%. In 77% of patients (n = 23), the most affected hemisphere was lateralized, and electrodes were placed ipsilateral to the most severely injured hemisphere in all; in 13% (n = 4), both hemispheres were equally affected and both sampled; in the remaining 10% (n = 3), pathology was diffuse, and sampling was unilateral (two patients with diffuse aSAH without focal ICH and one TBI with diffuse axonal injury [DAI]). The temporal lobe was the most frequently affected region (43% of patients), followed by frontal (30%) and parietal (3%) injury, with diffuse involvement without lobe predilection in the remainder (e.g., in cases of meningoencephalitis, aSAH without ICH, and widespread DAI). Importantly, approximately 80% of the most affected lobes were directly sampled with electrodes, meaning that an electrode was specifically dedicated to targeting the lesioned lobe and sampled both perilesional areas and adjacent nonlesioned gyri of that lobe. When the most affected lobe was not directly sampled, electrode coverage was always within the same, most severely affected hemisphere.

3.2. Spatial sampling and coverage

3.2.1. Superficial and deep cortical sampling

All patients achieved convexity cortical coverage. Superficial frontal cortex was sampled in all patients, parietal cortex in 70%, temporal cortex in 37%, and occipital cortex in 7% (Figure 3C). Nine patients (30%) additionally underwent deep implantation, targeting frontal (orbitofrontal cortex), temporal (hippocampus), parietal (posterior cingulate cortex), or insular structures (Figure 3D). At the electrode level, 72% of electrodes sampled cortical convexity exclusively, whereas 28% also sampled deep regions.

Electrode localization and density mapping demonstrated a distributed sampling pattern centered on frontal, parietal, and temporal cortices, with both superficial and deep coverage depending on technique (Figure 4). Density maps confirmed that coverage was not limited to a single gyral region but extended across multiple contiguous cortical territories within and across lobes. The relative density of intracranial electrode contacts mapping to specific anatomical cortical parcels is depicted in Figure 4 (for the absolute number of electrode contacts per atlas parcel, see Figure S1).

FIGURE 4.

Spatial distribution and anatomical coverage of intracranial electroencephalographic (EEG) sampling. The upper row shows kernel density estimation maps representing the spatial distribution of all intracranial EEG electrode contacts projected onto a common MNI305 cortical surface, visualized from dorsal, ventral, and lateral perspectives. Density values reflect the relative spatial sampling probability across subjects and implantation strategies. The middle row depicts electrode contact density mapped to anatomical cortical parcels using the Destrieux atlas, illustrating the extent of coverage across frontal, parietal, temporal, occipital, and insular regions, both superficial and deep. The lower row shows the individual electrode arrays and contact locations plotted on the same MNI305 surface, color‐coded by implantation strategy (open subdural, burr hole subdural, foramen ovale, and transcortical). Together, these representations demonstrate that multielectrode implantation strategies enable distributed, multilobar cortical sampling, including selective deep targets, beyond the focal coverage achieved by conventional intracranial EEG approaches in acute brain injury.

3.2.2. Cortical cluster coverage

Based on combined sampling of superficial and deep cortical regions across multiple lobes, four reproducible cortical clusters configurations emerged: frontal‐only (13%), frontoparietal (40%), frontotemporal (20%), and frontoparietotemporal (27%; Figure 3E).

Within multilobar configurations, frontal cortex consistently constituted the largest proportion of contacts (approximately 50%–70%), reflecting its surgical accessibility for coverage, whereas parietal and temporal contacts each contributed 20%–40% of within‐cluster sampling (Figure 3F). Individual patient trajectories leading to these distinct coverage patterns are illustrated in Figure S2 (Sankey plot).

3.3. Recording durability and monitoring characteristics

3.3.1. Timing and duration of monitoring

Combined iEEG–suEEG monitoring was initiated a mean of 20.6 h after surgery (SD = 16.6, 95% confidence interval [CI] = 13.7–27.5), with 88% of patients monitored within 24 h. After excluding patients without monitoring due to early death or withdrawal of care, total monitoring time was 361 patient‐days. Median monitoring duration was 13.8 days (interquartile range = 8.9–19.0), with a mean of 14.5 days (SD 7.6, 95% CI = 11.3–17.6).

3.3.2. Electrode contact survival

Electrode contact survival differed significantly by implantation strategy (log‐rank χ² = 19.53, df = 3, p = .0002; Figure S3; Kaplan–Meier curves do not account for intrapatient correlation of electrode contacts nested within subjects). After accounting for clustering by patient, implantation strategy remained independently associated with contact longevity (χ ² = 25.83, df = 3, p < .0001). Compared with open subdural electrodes, transcortical electrodes demonstrated a lower hazard of failure (hazard ratio [HR] = .30, 95% CI = .09–.98, p = .05), whereas burr hole subdural electrodes showed similar survival (HR = .87, 95% CI = .43–1.75, p = .69). No contact failures occurred in foramen ovale electrodes during the observation period (for Cox analysis excluding this group, see Figure S3). For the transcortical strategy, the Schoenfeld residual test suggested possible nonproportional hazards (p = .037); thus, the HR may vary over time and is interpreted cautiously given the modest number of patient clusters. At 15 days, contact survival probabilities were approximately 75%–80% for open and burr hole subdural electrodes, 94% for transcortical electrodes, and 100% for foramen ovale electrodes (table in Figure S3). These findings indicate that recordings can be maintained for prolonged ICU monitoring periods, with durability influenced by implantation approach.

3.4. Safety and feasibility

No intracranial hematomas attributable to electrode placement were detected on routine postoperative imaging or after explantation. Two patients experienced transient cerebrospinal fluid leakage at electrode exit sites, resolving with bedside suturing. Three cases of ventriculitis occurred during prolonged ICU stays and were considered possibly related but confounded by multiple, more plausible alternative risk factors (multiple external ventricular drain [EVD] and lumbar drain procedures in all patients, use of non‐antibiotic‐coated EVDs, prolonged skull base cerebrospinal fluid leakage; Table 1). No wound infections or electrode dislodgements occurred.

TABLE 1.

Adverse events.

Patient	Adverse event	Day of onset	Age, years	Etiology	Implantation	Causal probability	(Alternative) cause	Management	Outcome
5	Transient CSF leakage	1	64	CNSi	Burr hole subdural depth electrode	Certain	Electrode entered superficial arachnoid cyst.	Extra bedside stitch	Resolved
12	Transient CSF leakage	1	50	TBI	Open subdural + FO electrode	Certain	Tunneled to occipital area instead of vertex.	Extra bedside stitch	Resolved
	Ventriculitis	16				Possible, but alternative causes more probable	Patient underwent prolonged external drainage. The EVD lacked antibiotic coating, elevating ventriculitis risk. Additionally, the patient experienced prolonged external CFS leakage in the first week due to a skull‐base fracture (high risk for meningitis/ventriculitis).	Antibiotics	Resolved
8	Ventriculitis	8	72	aSAH	Burr hole subdural depth electrodes	Possible, but alternative causes more probable	Patient underwent 2 EVD procedures (high risk for ventriculitis). Both EVDs lacked antibiotic coating (higher ventriculitis risk). Additionally, the patient received an ELD before developing ventriculitis.	Antibiotics	Resolved
18	Ventriculitis	18	31	TBI	Stereotactic transcortical depth electrodes	Possible, but alternative cause more probable	Patient underwent 2 EVD procedures (high risk for ventriculitis) along with prolonged drainage.	Antibiotics	Resolved

Note: Overview of adverse events, probability of causation by electrode or procedure, causes of the event, management, and outcome. In the two patients who experienced transient CSF leakage from the electrode's tunneled exit point, concurrent leakage also occurred from the EVD's exit point. This was promptly resolved by placing a bedside stitch. Following these events, we adopted a tunneling needle with a smaller diameter, after which no further leaks were observed.

Abbreviations: aSAH, aneurysmal subarachnoid hemorrhage; CNSi, central nervous system infection; CSF, cerebrospinal fluid; ELD, external lumbar drain; EVD, external ventricular drain; FO, foramen ovale; TBI, traumatic brain injury.

Across 30 patients, 64 electrodes, and 361 patient‐monitoring days, per‐patient risk and per‐electrode risk were 13.3% and 7.8% for any adverse event, and 6.7% and 3.1% for causally related events, respectively. Normalized to exposure, this corresponded to .24 causally related events per 100 electrode‐days (95% CI = .03–.87) and .60 (95% CI = .07–2.19) per 100 patient‐days.

Implantation success was high; 96% of intended electrodes were implanted, of which 95% followed the planned trajectory (see Methods S1 for reasons to abstain from implantation or deviate from planned trajectory). All electrodes were removed successfully at the bedside. Additional operative time was minimal for subdural approaches (5–10 min) and longer for navigated deep implants (30 min).

3.5. Electrophysiological findings and clinical impact

Intracranial recordings frequently revealed clinically relevant epileptiform activity and influenced treatment decisions. Among 25 monitored patients (five without EEG, due to early death or withdrawal in four and early recovery in one), one patient (4%) had a history of epilepsy and six (24%) had an acute symptomatic convulsive seizure after injury impact. Epileptiform activity was observed in 12 (48%) on iEEG and 11 (44%) on suEEG, with 15 (60%) showing epileptiform patterns on either modality (Table 2). Sporadic epileptiform discharges (SEDs) were noted in four (16%) patients, seizures were observed in five (20%), and RPPs in 14 (56%). Of the 15 patients with epileptiform activity, 12 (80%) received antiseizure medications, including seven (47%) treated as nonconvulsive status epilepticus (NCSE). Three patients with only a low prevalence of SEDs or RPPs were not treated. Of the seven patients urgently treated according to an NCSE protocol, four (57%) were based on iEEG findings, one (14%) based on suEEG, and two (29%) equally based on both methods. iEEG findings drove initiation or escalation of antiseizure therapy in eight of 15 patients (53%; 32% of all monitored patients), whereas suEEG was the main modality for guiding therapy in two patients (13%; 8% of all monitored patients). In two patients (13%; 8% of all monitored patients), treatment was equally based on both methods. When treatment initiation and escalation were based on iEEG and not suEEG, this was typically due to a higher epileptic pattern burden, more severe patterns, or missed abnormalities such as seizures or RPPs, reflecting a more aggressive disease course. Detailed formal yield and blinded comparative analyses are not reported here and are the focus of a separate study.

TABLE 2.

Electrophysiological findings and clinical impact.

Patient	iEEG approach	iEEG findings	suEEG findings	Modality used for treatment initiation and escalation
1	Open subdural	—	—	—
2	Open subdural + FO	SED, RPP, ESz	SED, RPP, ESz	NCSE treatment based on iEEG
3	Open subdural + FO	—	—	—
4	Burr hole subdural + transcortical	—	—	—
5	Burr hole subdural	RPP	—	RPP treatment based on iEEG
6	Burr hole + open subdural	RPP	RPP	NCSE treatment based on iEEG
7	Open subdural	—	—	—
8	Burr hole subdural	RPP, ESz	RPP	NCSE treatment based on iEEG
9	Open subdural	—	—	—
10	Open subdural	—	—	—
11	Burr hole subdural + FO	SED	—	—
12	Open subdural + FO	—	RPP	RPP treatment based on suEEG; no escalation since iEEG negative
13	Burr hole subdural + transcortical	—	RPP	—
14	Open subdural	—	RPP	—
15	Burr hole subdural	RPP, ESz	RPP, ESz	NCSE treatment based on both modalities
16	Burr hole subdural	—	—	—
17	Burr hole subdural	RPP	—	RPP treatment based on iEEG
18	Burr hole subdural + transcortical	RPP, ESz	—	RPP–ESz treatment based on iEEG
19	Burr hole + open subdural	RPP	RPP	NCSE treatment based on both modalities
20	Open subdural	—	—	—
21	Burr hole subdural	RPP	RPP, ESz	NCSE treatment based on suEEG
22	Open subdural	—	—	—
23	Burr hole subdural + transcortical	—	—	—
24	Transcortical	SED, RPP	RPP	RPP treatment based on iEEG
25	Burr hole subdural	SED, RPP	SED	RPP treatment based on iEEG

Note: Descriptive electrophysiological findings and impact on management in 25 patients with EEG data. Definitions of epileptiform activity were based on the American Clinical Neurophysiology Society 2021 terminology.

Abbreviations: EEG, electroencephalographic; ESz, electrographic seizures; FO, foramen ovale; iEEG, intracranial EEG; NCSE, nonconvulsive status epilepticus; RPP, rhythmic and periodic patterns; SED, sporadic epileptiform discharges; suEEG, surface EEG.

Representative traces showing iEEG versus suEEG concordance are depicted for each implantation strategy in Figure S4.

3.6. Comparison with existing techniques

Figure S5 summarizes a qualitative, multidimensional comparison between the proposed strategies and conventional approaches. No single implantation strategy optimizes all dimensions simultaneously.

4. DISCUSSION

In this prospective cohort, we evaluated four iEEG strategies designed to expand spatial sampling options in patients with acute brain injury. The motivation for this work stems from converging observations that (1) most epileptic events in critically ill patients are nonconvulsive and require EEG for detection, (2) scalp EEG has limited sensitivity for focal or deep epileptic sources, and (3) seizures and RPPs are common and may contribute to secondary brain injury and worse outcomes. Against this background, the present study is the first to investigate whether intracranial sampling strategies are technically feasible, durable, and safe, and whether they provide reproducible distributed anatomical coverage, within constraints of acute neurosurgical and ICU care.

4.1. Spatial sampling and coverage in ICU iEEG

The concept of sampling and coverage in critical care invasive EEG differs from the presurgical evaluation in refractory epilepsy. Instead of defining the epileptogenic network through localization of seizure onset areas and network involvement, detecting whether epileptic patterns are present could be the main goal of ICU EEG. Experimental and clinical data suggest that seizures and periodic discharges may arise from hyperexcitable, perilesional cortex, but the concept of acute epileptogenesis is not well understood. ¹⁷ , ²⁴ , ²⁸ , ²⁹ , ³³ Injuries can be focal, multifocal, or diffuse, and epileptic activity is expected to exhibit heterogenous characteristics, both regionally within patients and across patients. ¹ , ² , ²⁷ , ³⁴ Based on these observations, it was hypothesized that multifocal, poorly synchronized sources of epileptic activity could potentially account for ictal dynamics undetectable on scalp EEG. ²⁹ Current invasive methods in acute brain injury involve either a single subdural strip positioned perilesionally during craniotomy or a short intracortical ("mini") depth electrode inserted through a burr hole or bolt, usually in a standardized location within the frontal cortex. ⁸ , ¹² , ²⁵ , ²⁶ , ²⁸ , ²⁹ Although these strategies have suggested increased sensitivity for seizure and RPP detection, they provide highly focal recordings that are vulnerable to spatial undersampling and are confined to specific settings of brain injury. ²⁷ Prior studies employing these methods have documented epileptic activity that was undetected on iEEG, indicating that such events can occur beyond the coverage area and thus sampled field of the invasive electrodes. ²⁵ Moreover, seizures and RPPs detected on iEEG can be confined to a limited subset of electrode contacts. ²⁶ , ²⁸ Because the primary objective is detection, this creates uncertainty about how frequently a negative iEEG result truly signifies the absence of seizures. In settings with (multi)focal or diffuse lesions, expanding lesion‐targeted anatomical coverage could inherently offer a greater likelihood of detecting localized epileptic activity.

From this perspective, we designed a combination of subdural or stereotactic depth electrode‐based strategies that allowed expansion of cortical coverage while remaining compatible with acute workflows. By combining approaches for convexity and deep sampling, we achieved reproducible coverage of lesioned frontal, parietal, and temporal cortical regions. These approaches obviated the need for additional burr holes, craniotomies, or second procedures. With the perilesional hypothesis in mind, electrodes were placed ipsilateral to the most severely injured hemisphere and targeted toward the most affected lobes whenever feasible. As a result, when lateralized, the most damaged hemispheres were always sampled, and in nearly 80% the most damaged lobe was directly covered. Across patients, this approach yielded consistent frontoparietal, frontotemporal, and frontoparietotemporal cortical cluster coverage patterns. Importantly, “coverage” should be interpreted as anatomical access to regions that might potentially harbor epileptic patterns, as opposed to evidence of electrophysiological representativeness.

In this context, the objective of the present study was to broaden the opportunity for detection, that is, to enhance the probability of recording epileptic activity arising in anatomically varied or deeper brain regions, while refraining from claiming superior accuracy over established methods. This study offers a framework that can be tailored for investigating epileptic dynamics in acute brain injury. ²⁷ We demonstrate that adopting this iEEG strategy impacted ICU management; iEEG guided treatment initiation or escalation in 32% of all monitored patients and in 57% of those considered status epilepticus.

4.2. Depth electrode‐based ECoG

We intentionally used depth electrodes (thin linear arrays [1 mm] with relatively high contact density and large recording lengths), which are classically used for intraparenchymal trajectories, to enable (1) burr hole compatibility, (2) reproducible linear sampling over longer distances, and (3) straightforward removal. The “depth‐electrode ECoG” concept complements existing techniques that use subdural strips after craniotomy or short depth electrodes for bolt‐based or burr hole monitoring, while shifting the emphasis toward minimizing invasiveness while increasing cortical coverage. The implantation of depth electrodes in the subdural space uses existing surgical corridors without penetrating brain tissue. It can therefore be considered as a less invasive intracranial approach that can be easily implemented.

4.3. Recording durability and feasibility in the ICU environment

A key requirement for invasive EEG in acute brain injury is the ability to maintain stable recordings in a challenging environment, often over prolonged ICU courses. Across all implantation strategies, we observed robust long‐term recording performance, with median monitoring durations approaching 2 weeks, surpassing current monitoring periods in the ICU. Although electrode contact survival differed by implantation type, which is likely to reflect mechanical stress and electrode fixation differences, high‐quality signals were maintained across all strategies for durations relevant to acute seizure and secondary injury monitoring.

Feasibility was further supported by high implantation success rates, short additional operative time, rapid initiation of monitoring, and uncomplicated bedside explantation. These features are essential for generalizing invasive EEG beyond highly selected cases and into routine practice.

4.4. Safety in context of existing invasive EEG literature

The safety profile observed in this cohort compares favorably with prior invasive EEG studies in acute brain injury. ³ , ²⁶ No hemorrhagic complications attributable to electrode placement were observed. Although 13% of patients had adverse events, the rate of complications causally attributed to electrode implantation was low (per‐patient and per‐electrode risk of 6.7% and 3.1%, respectively), with low severity. The occurrence of ventriculitis seemed more attributable to the long‐term use of multiple non‐antibiotic‐coated ventricular drains, a known risk factor. The complication rate was slightly higher compared to large stereo‐EEG safety series reporting complications rates of 1%–4% and more comparable to rates seen in subdural monitoring in refractory epilepsy populations (approximately 9%). ³⁵ , ³⁶ These differences reflect the variable risk profiles of patients with severe brain injury undergoing acute and extensive surgery. Our data, however, support the notion that invasive EEG can be safely extended to critically ill populations. ³⁵

4.5. Limitations

First, this study does not independently show improved detection outcomes, superiority to other techniques, or improved clinical outcomes. The proposed strategy should be viewed as a conceptual framework allowing further study and advancement in the field of acute epileptogenesis. Next, it was conducted at a single center with a modest sample size, and some implantation strategies, particularly foramen ovale electrodes, were used in relatively few patients. The modest number of patient clusters may affect confidence intervals and p‐values for hazard ratios, which should be interpreted with caution. We acknowledge that the use of foramen ovale electrodes and transcortical electrodes based on stereotactic techniques requires a higher level of expertise and is not always easy to implement compared to subdural approaches. Electrode localization relied on CT‐based registration, which introduces some spatial imprecision, although common in critically ill populations.

4.6. Future directions

The question moving forward is how the spatial sampling strategy influences diagnostic yield and how invasive strategies should be adapted to patient‐ and pathology‐specific characteristics. Planned analyses in this cohort will evaluate incremental diagnostic yield compared with suEEG, compare epileptic patterns across electrodes, and integrate spatial coverage metrics into analyses of seizure and RPP detection in relation to lesion topography. By making distributed intracranial sampling technically feasible, durable, and safe in the ICU, this work provides mainly a methodological foundation for future research on epileptic dynamics in acute brain injury.

5. CONCLUSIONS

Four complementary, multielectrode strategies extend iEEG coverage in the ICU population with acute brain injury, with acceptable risk and durable long‐term recordings. These methods are feasible, implementable, and generalizable to any setting of acute brain injury where invasive neuromonitoring is standard of care.

AUTHOR CONTRIBUTIONS

Conceptualization: Steven Smeijers and Tom Theys. Data curation: Steven Smeijers and Anaïs Van Hoylandt. Formal analysis: Steven Smeijers and Mariella Segreti. Funding acquisition: Steven Smeijers and Tom Theys. Investigation: Steven Smeijers and Anaïs Van Hoylandt. Methodology: Steven Smeijers, Tom Theys, Jeroen Gijs, Lauranne Scheldeman, Bart Depreitere, and Johannes van Loon. Project administration: Steven Smeijers and Anaïs Van Hoylandt. Resources: Johannes van Loon, Bart Depreitere, and Tom Theys. Software: Steven Smeijers and Mariella Segreti. Supervision: Johannes van Loon, Bart Depreitere, Lauranne Scheldeman, and Tom Theys. Validation: Jeroen Gijs, Lauranne Scheldeman, Bart Depreitere, and Tom Theys. Visualization: Steven Smeijers and Mariella Segreti. Writing—original draft: Steven Smeijers. Writing—review & editing: Mariella Segreti, Jeroen Gijs, Johannes van Loon, Lauranne Scheldeman, Bart Depreitere, and Tom Theys. Authorship criteria are based on those outlined by the International Committee of Medical Journal Editors Uniform Requirements for Manuscripts Submitted to Biomedical Publications.

FUNDING INFORMATION

S.S. received a research grant from Research Foundation Flanders (FWO; PhD fellowship fundamental research—11K8923N). This work was further supported by internal KOOR funding from University Hospitals Leuven, an FWO grant supporting T.T. (G0B6422N), and a KU Leuven grant C14/22/134. T.T. is a senior clinical investigator of FWO (1830722N).

CONFLICT OF INTEREST STATEMENT

None of the authors has any conflict of interest to disclose.

ETHICS STATEMENT

This study was approved by the ethics committee of UZ Leuven—KU Leuven (s67062), and written consent was obtained by the patients' representatives prior to study enrollment. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

CLINICAL TRIAL REGISTRATION

The results are part of a prospective study on intracranial EEG in brain injury (ClinicalTrials.gov ID NCT06426615—CONSCIUS study).

Supporting information

Figure S1. Intracranial contacts per anatomical parcel.

Figure S2. Sankey plot.

Figure S3. Electrode contact survival by implantation strategy. Kaplan–Meier curves depict survival of individual intracranial electroencephalographic (EEG) contacts over time, stratified by implantation technique (open subdural, burr hole subdural, foramen ovale, and transcortical). Note that the Kaplan–Meier curves do not account for intrapatient correlation of electrode contacts nested within patients. Contact failure was defined as loss of an interpretable EEG signal due to impedance increase or dysfunction. Shaded areas represent 95% confidence intervals. Survival differed significantly between implantation strategies (log‐rank p = .0002). Foramen ovale and transcortical electrodes demonstrated high long‐term contact survival, whereas subdural depth electrodes showed some gradual degrading over time. Despite these differences, all strategies provided robust recording durability compatible with prolonged intensive care unit monitoring. The accompanying table reports the number of contacts at risk and estimated contact survival probabilities at predefined time points. N, number.

Figure S4. Representative electroencephalographic (EEG) traces. (A) Open subdural and foramen ovale technique, showing scalp EEG and concomitant intracranial EEG (iEEG) with one open subdural electrode (14 contacts) and one foramen ovale electrode (four contacts). Surface EEG shows lateralized rhythmic delta activity, whereas iEEG shows periodic discharges and start of an electrographic seizure. Surface EEG is contaminated by artifact. (B) Burr hole subdural technique, showing scalp EEG and concomitant iEEG with two subdural burr hole electrodes (14 contacts each). Surface EEG and iEEG show a burst–suppression pattern. (C) Transcortical technique, showing scalp EEG and concomitant iEEG with two transcortical electrodes (nine contacts each). iEEG shows periodic discharges on the deepest contacts of electrode B (B1‐3) invisible to scalp EEG.

Figure S5. Comparative overview of invasive electroencephalographic (EEG) techniques for acute brain injury, showing conceptual comparison of the four proposed multielectrode implantation strategies (open subdural, burr hole subdural, foramen ovale, and transcortical) versus commonly used focal intracranial EEG approaches (subdural strip and intracortical depth electrode). Techniques are compared across multiple domains, including recording modality, electrode characteristics, invasiveness, surgical setting, fixation, implantation technique, cortical and deep coverage, adaptability to patient‐specific anatomy and pathology, approximate additional surgical time, contact level durability, complication profile, and typical clinical application. Colors reflect relative strengths and limitations within each domain, intended as a qualitative synthesis. The figure highlights tradeoffs between spatial coverage, invasiveness, durability, and feasibility. c, contacts; d, days; ECoG, electrocorticography; FO, foramen ovale; w, weeks.

DATA AVAILABILITY STATEMENT

There are no data that can be shared systematically.