Stereo-EEG & subdural grid planning for epilepsy surgery — from non-invasive hypothesis to trajectory design, implantation, and functional mapping
Before a single electrode touches the skull, every candidate for invasive epilepsy monitoring passes through weeks of non-invasive evaluation. Scalp video-EEG captures the seizure semiology and surface ictal rhythm; high-resolution 3T MRI hunts for a structural lesion (focal cortical dysplasia, mesial temporal sclerosis, cavernoma); interictal PET and ictal SPECT reveal regional metabolic and perfusion abnormalities. Case conference synthesizes all of it into a single working hypothesis of where the seizure-onset zone lies — and that hypothesis, not the raw data alone, determines exactly where invasive electrodes will be placed.
Long-term video-EEG monitoring is the backbone of pre-surgical evaluation:
• Patients are admitted to an epilepsy monitoring unit (EMU); anti-seizure medications are gradually tapered to provoke habitual seizures under continuous video and 21+ channel scalp EEG recording • Semiology (the observable clinical behavior of a seizure — automatisms, head version, dystonic posturing, vocalization) is analyzed frame-by-frame and correlated with the evolving ictal EEG rhythm • Interictal epileptiform discharges (spikes, sharp waves) are mapped for lateralizing and localizing value, though their yield is coarser than ictal recordings • Ictal onset pattern on scalp EEG (rhythmic theta, low-voltage fast activity, or attenuation) provides a first, often imprecise, localization — scalp electrodes are 10–20+ mm from cortex and heavily filtered by skull and scalp
Scalp EEG alone is frequently insufficient for localization to within a resectable region — especially for deep, bilateral, or multifocal generators — which is precisely the population referred onward to invasive intracranial evaluation.
Three imaging modalities are combined to triangulate the epileptogenic zone before any invasive electrode is placed:
• High-resolution structural MRI (3T, dedicated epilepsy protocol with thin-slice T1, FLAIR, and susceptibility sequences) is scrutinized for a lesion: focal cortical dysplasia, hippocampal sclerosis, low-grade tumor, vascular malformation, or encephalomalacia. A visible, concordant lesion dramatically raises surgical success rates. • Interictal FDG-PET detects regional hypometabolism, often extending beyond the structural abnormality and helping delineate the broader epileptogenic network, particularly valuable when MRI is normal ("MRI-negative" epilepsy). • Ictal-interictal SPECT subtraction (SISCOM) captures the transient hyperperfusion that accompanies seizure onset by injecting a radiotracer within seconds of ictal EEG onset — a logistically demanding but highly localizing study.
MEG (magnetoencephalography) and functional MRI (language/motor mapping, resting-state connectivity) are added at many centers to further refine hypotheses and identify eloquent cortex to be protected later.
All non-invasive data converges in a formal epilepsy surgery conference — epileptologists, neurosurgeons, neuroradiologists, neuropsychologists, and biomedical engineers review the case together:
• Concordant data (semiology + scalp EEG + MRI lesion + PET/SPECT all point to the same region): the patient may proceed directly to resection, or invasive EEG may still be used to confirm boundaries near eloquent cortex • Discordant data (conflicting localization across modalities, non-lesional MRI, suspected bilateral or multifocal onset, or a lesion near eloquent cortex): invasive intracranial EEG is required to directly sample the candidate network • The output of this conference is not a diagnosis but an explicit, falsifiable anatomo-electro-clinical hypothesis: a ranked list of candidate onset regions and network nodes that the invasive electrode plan must be built to test
The invasive electrode plan is only as good as the hypothesis behind it — every electrode placed is essentially a pre-registered test of a specific anatomical prediction, not a blind exploratory search of the whole brain.
Once a hypothesis exists, the team must choose the tool that best tests it. Stereo-EEG (SEEG) uses thin, multi-contact depth electrodes inserted stereotactically to sample deep, bilateral, or multiple discontinuous regions with minimal surgical footprint. Subdural grid and strip electrodes are flat arrays of contacts laid directly on an exposed cortical surface via craniotomy, offering dense, contiguous 2D mapping of an accessible convexity — at the cost of open surgery and inability to reach deep or bilateral structures easily.
SEEG electrodes are thin (0.8mm diameter) multi-contact cylindrical probes, each carrying 8 to 18 platinum-iridium contacts spaced 3.5–5mm apart along the shaft:
• Inserted through small twist-drill or burr holes under stereotactic (frame-based or robotic) guidance, following a straight trajectory from scalp entry to one or more deep or superficial targets • A single oblique trajectory can sample multiple structures along its path (e.g., orbitofrontal cortex → insula → amygdala → hippocampus), which is the central efficiency advantage of SEEG • 8–16 electrodes per patient is typical, allowing bilateral and multilobar sampling in a single implant with minimal craniotomy footprint (band-aid sized entry wounds instead of an open flap) • Best suited for: deep foci (insula, cingulate, orbitofrontal, mesial temporal structures), bilateral or multifocal hypotheses, and re-operations where prior craniotomy scarring complicates subdural placement
Subdural electrodes are flexible silicone sheets embedding flat platinum discs (typically 4mm diameter, 2.3mm exposed, spaced 1cm apart in grids up to 8×8 = 64 contacts, or in 1×4 to 1×8 linear strips):
• Requires a craniotomy — an open bone flap — to place the array directly on the pial (subdural) surface, giving dense, contiguous, high-fidelity mapping of an entire accessible gyral convexity • Superior for delineating the precise 2D boundary between epileptogenic and eloquent cortex on an accessible lateral convexity, because contact geometry is fixed and evenly spaced across a known cortical patch • Cannot easily reach mesial, deep, interhemispheric, or bilateral structures without extensive additional craniotomy — geometrically limited to the exposed surface • Historically the dominant modality; now reserved mainly for cases needing dense convexity mapping adjacent to eloquent cortex, or where SEEG trajectories are contraindicated by vascular anatomy
The choice is rarely absolute — many centers combine both:
• SEEG is favored when: the hypothesis involves deep structures (insula, cingulate gyrus, mesial temporal lobe), bilateral independent foci, multiple discontinuous lobes, or a small/band-aid-scar surgical footprint is clinically preferred • Subdural grids are favored when: the hypothesis is confined to an accessible lateral convexity immediately adjacent to eloquent cortex (e.g., peri-Rolandic epilepsy) and dense, evenly-spaced 2D mapping plus direct cortical stimulation is the priority • Hybrid implants (SEEG depth electrodes plus a subdural strip) are used when deep and superficial hypotheses coexist • A key driver of the field-wide shift toward SEEG over the past decade has been its markedly lower hemorrhagic and infectious complication rate relative to open subdural craniotomy, alongside its superior reach into deep, multifocal, and bilateral networks
For SEEG in particular, every one of the 8 to 16 planned trajectories must be individually engineered: a straight line from a safe scalp entry point, through sulcal and gyral tissue, to a hypothesized network node — without crossing a single vessel visible on angiographic imaging. Purpose-built trajectory planning software fuses MRI, CT angiography, and venography into one 3D model and lets the surgical team optimize dozens of candidate paths simultaneously, scoring each for vascular risk before any electrode is inserted.
Trajectory planning begins by co-registering multiple imaging studies into a single 3D stereotactic space:
• High-resolution T1 MRI provides the anatomical target coordinates and gyral/sulcal surface for entry point selection • CT angiography (CTA) and CT venography (CTV), or contrast-enhanced MR angiography/venography, delineate every artery and vein down to 1mm caliber — including cortical veins, which are just as critical to avoid as arteries • All studies are fused into one coordinate system in dedicated planning software (e.g., ROSA, Neurolocate, StealthStation, EpiNav), rendering a 3D vascular "no-fly zone" model around the entire cranium and brain parenchyma
Accurate fusion is the foundation of the entire planning process: a registration error of even 1–2mm can shift a computed trajectory directly into a vessel that appeared clear in the plan.
For each hypothesized target, planning software proposes and scores many candidate entry points and trajectory angles:
• A minimum vascular clearance (typically ≥3mm from any vessel along the entire trajectory length) is enforced as a hard constraint • Trajectories are penalized for steep angles relative to the skull (risking electrode slippage or fracture), for crossing sulcal folds at oblique angles (increasing vessel-crossing probability), and for excessive length (increasing cumulative risk) • Multi-target trajectories are specifically optimized to pass through several hypothesized structures along one line (e.g., a single oblique insular-opercular-temporal trajectory), maximizing information yield per electrode • The output is a ranked shortlist of candidate trajectories per target; the implanting neurosurgeon makes the final selection, balancing the algorithm's vascular safety score against anatomical and functional judgment
Once trajectories are finalized, one of three guidance systems executes them:
• Frame-based stereotaxy: a rigid head frame fixed to the skull provides a physical coordinate reference; historically the gold standard for sub-millimeter accuracy, but slower for multi-electrode implants • Frameless (optical or electromagnetic) navigation: skull-fiducial or surface registration replaces the frame, trading a small amount of accuracy for substantially faster multi-trajectory workflows • Robotic stereotaxy (e.g., ROSA, Neuromate): a robotic arm automatically aligns to each pre-planned trajectory in sequence, holding a guide tube through which the surgeon drills and inserts the electrode — dramatically reducing per-electrode setup time for 12–16 electrode implants while maintaining sub-millimeter entry-point accuracy
A single misregistered or poorly scored trajectory is the single most common preventable cause of intracranial hemorrhage in SEEG — which is why modern centers treat trajectory planning as a formal, auditable step with explicit vascular clearance thresholds rather than surgeon intuition alone.
Implantation translates the digital trajectory plan into physical reality. For SEEG, this means dozens of individually drilled twist-drill or burr holes, each guided by frame, frameless, or robotic stereotaxy to insert a depth electrode along its pre-computed path. For subdural grids, a craniotomy exposes the cortical surface directly, and flexible electrode sheets are laid by hand under direct visualization before the bone flap is replaced.
Depth electrode placement follows a tightly choreographed sequence per trajectory:
1. The patient's head is fixed (frame) or registered (frameless/robotic) to the pre-operative imaging 2. For robotic cases, the arm aligns a guide tube precisely along the planned trajectory at the scalp entry point 3. A twist-drill or small burr hole is made through the skull; a bolt or anchor is secured to fix the electrode in place post-insertion 4. The depth electrode is advanced along the guide to the pre-computed target depth, with real-time or immediate post-insertion imaging confirming contact positions match the plan 5. This is repeated sequentially for every one of the 8–16 planned electrodes, typically completed in a single 3–6 hour procedure 6. A post-implantation CT (fused back onto the pre-operative MRI) formally verifies every contact's anatomical location before recording begins — any trajectory deviation is documented and factored into subsequent interpretation
Subdural electrode placement is an open surgical procedure:
1. A craniotomy is planned and sized to expose the full cortical region implicated by the hypothesis, typically guided by neuronavigation 2. The dura is opened, and flexible silicone grid and/or strip arrays are laid directly onto the pial surface, conforming to the gyral anatomy under direct visualization 3. Strips can additionally be slid beneath the dura beyond the craniotomy margin to extend coverage without further bone removal 4. Electrode leads are tunneled subcutaneously to exit sites away from the incision, then connected to the recording system 5. The bone flap is replaced and secured, and the patient recovers before transfer to the epilepsy monitoring unit for the recording phase 6. Because the dura and bone flap are temporarily left more mobile to accommodate brain swelling from the implanted hardware, subdural implantation carries a distinct risk profile — more prone to mass-effect complications, less prone to the deep-trajectory hemorrhage risk that dominates SEEG
Both modalities require vigilant peri-operative monitoring:
• Immediate post-implant CT is mandatory in both approaches to detect occult hemorrhage before it becomes symptomatic • Prophylactic antibiotics and careful wound care reduce infection risk, which rises with implant duration (both modalities typically remain in place 1–3 weeks during monitoring) • Neurological checks are performed after every trajectory in SEEG and immediately post-craniotomy for subdural grids, watching for any new deficit suggesting vascular injury • Modern high-volume SEEG centers report symptomatic hemorrhage rates of roughly 1–3%, meaningfully lower than historical subdural grid series, which contributed heavily to SEEG's rise as the default modality worldwide
With electrodes safely in place, the patient enters an epilepsy monitoring unit for continuous intracranial video-EEG recording — anti-seizure medication remains tapered until enough habitual seizures are captured directly from the implanted contacts to define the seizure-onset zone with millimeter precision. Between seizures, electrical stimulation mapping is performed contact-by-contact to identify eloquent cortex — motor, language, and sensory regions that must be preserved in any subsequent resection.
Once implanted, every electrode contact is connected to a high-channel-count recording system (often 128–256 channels) providing continuous video-synchronized intracranial EEG:
• Signal fidelity is dramatically higher than scalp recording — contacts are millimeters from the generating cortex, unfiltered by skull and scalp, revealing high-frequency oscillations (ripples, fast ripples) invisible on scalp EEG that are increasingly recognized as markers of the epileptogenic zone • Anti-seizure medication remains reduced or withheld to provoke habitual seizures within the monitoring window • Each captured seizure is analyzed for its precise ictal onset pattern (which contacts activate first, and in what sequence) and its propagation pathway across the implanted network — directly testing the pre-implantation hypothesis • Interictal recordings between seizures are also analyzed: the spatial extent of interictal spiking and high-frequency oscillations helps refine the boundaries of the presumed epileptogenic zone
Between spontaneous seizures, the clinical team performs functional mapping by delivering controlled electrical current through pairs of implanted contacts:
• Bipolar stimulation (current passed between two adjacent contacts) is delivered in a stepwise fashion, typically starting at 1mA and increasing to a clinical response or an afterdischarge threshold, commonly up to 10–15mA at 50Hz for 1–5 second trains • Motor cortex stimulation produces observable movement or a reported sensation in the corresponding body part; language cortex stimulation during a naming or reading task can produce speech arrest or naming errors; visual cortex stimulation can elicit phosphenes • Every contact pair's response (or lack of one) is logged onto the same 3D reconstruction used for trajectory planning, building a functional map directly overlaid on the anatomical implant • This mapping is the definitive, patient-specific alternative to population-average functional atlases — critical because eloquent cortex location varies meaningfully between individuals, especially near lesions that can displace normal function to adjacent tissue
The entire purpose of the implant converges here: the seizure-onset zone identified from recorded seizures and the eloquent cortex identified from stimulation mapping are overlaid on the same anatomical model, directly defining the resection that can stop seizures while preserving function.
The monitoring phase concludes with explanation of the electrodes and a second multidisciplinary conference:
• The seizure-onset zone (from ictal recordings), the irritative zone (from interictal spikes/HFOs), and the eloquent cortex map (from stimulation) are combined into a single 3D surgical plan • Where the seizure-onset zone and eloquent cortex overlap, the team must weigh seizure freedom against functional preservation — sometimes favoring a smaller, staged, or laser-ablation resection over an aggressive one • Electrodes are then surgically removed, typically at the start of the definitive resective or ablative procedure, having served their sole purpose of localizing what standard imaging alone could not resolve • Modern series report seizure freedom in roughly 50–70% of patients who proceed to resection after a well-localized SEEG or subdural evaluation — underscoring how much the quality of this entire planning pipeline, from hypothesis to stimulation map, determines the surgical outcome