Before the operation,
understand the evidence.

The decision is bigger than “which side?”

A patient arrives with seizures that continue despite treatment. The team must answer several questions: Are all the events epileptic? Is there a treatable focal source or network? What evidence supports that hypothesis? What functions would be at risk? Does the expected benefit justify the burden and risk for this particular person?

This guide follows the reasoning behind those questions. It is written for an engineering student observing an epilepsy service, not as instructions for evaluating a patient independently. The main example is focal epilepsy; childhood developmental disorders, generalized epilepsies, and hemispheric syndromes can follow different pathways.

Drug resistance prompts specialist evaluation—not automatic surgery.

The ILAE definition requires failure to achieve sustained seizure freedom after adequate trials of two tolerated, appropriately selected and used antiseizure medication regimens, alone or in combination. Merely counting prescriptions is insufficient. Clinicians reconstruct the drug, indication, dose and duration, adherence, response, and reason for stopping. An inadequately documented or unsuitable trial may be uninformative rather than evidence of resistance. [1]

The 2022 ILAE expert consensus supports prompt surgical-evaluation referral once drug resistance is established, with age and clinical circumstances addressed in its recommendations. Referral also creates an opportunity to reconsider diagnosis and optimize other care. It should not be delayed simply because surgery feels like a “last resort.” [2] NICE likewise recommends referral for assessment in drug-resistant epilepsy, including when MRI shows no abnormality; selected structural lesions can justify earlier specialist assessment. These recommendations support evaluation, not an obligation to undergo an operation. [4]

A team builds the case with the patient.

Presurgical assessment brings together neurological, cognitive, psychological, and social information. The roles below are a practical orientation; titles and responsibilities differ between centers. [7]

The multidisciplinary discussion integrates these views. It does not replace informed, shared decision-making with the patient.

A branching pathway, not a conveyor belt.

Phase 1 establishes a noninvasive hypothesis. Selected patients need Phase 2 intracranial recording to resolve an actionable uncertainty. Others proceed without it, or do not become surgical candidates. [3]

Imaging order: structural MRI usually provides the anatomical foundation. Additional modalities are selected for the question, availability, and existing findings—not because everybody must complete MRI → PET → SPECT → MEG → SEEG. Functional-risk testing may proceed alongside localization. Re-review of earlier data is common. [26]

Six maps that should not be collapsed into one.

Conceptual distinctions summarized from clinical presurgical literature. Terminology and network models continue to evolve. [5] [23]

Lateralization asks which hemisphere; localization asks which region or network. A proposed surgical site and its boundaries require substantially more information than a left/right answer.

The story is part of the measurement.

Clinicians ask about the earliest remembered sensation, awareness, movements, speech, duration, recovery, triggers, and whether there are several event types. Witness accounts and available video can supplement an incomplete memory. Examination and earlier investigations provide additional context. Diagnostic tests support this assessment rather than replacing it. [6]

Different measurements answer different questions.

EEG and MEG measure electrophysiology; MRI describes anatomy or, with specialized acquisitions, indirect functional signals and diffusion. PET and SPECT use radiotracers. They are not interchangeable pictures of “the epilepsy.” The cards below are a reading guide, not a required test list.

Wait for events. Record continuously. Review selectively.

In an epilepsy monitoring unit (EMU), scalp EEG and synchronized video capture the event’s electrical and behavioral evolution. Staff document clinical changes while maintaining safety and signal quality. Johns Hopkins describes continuous recording and a usual stay of three to seven days in its service. That is a center example, not a guaranteed admission length. [16]

Capturing habitual events helps determine whether they are epileptic and how their clinical sequence relates to EEG changes. Specialist teams may adjust medication during monitored admission to increase the chance of recording events; this is supervised hospital practice, not something patients or students should reproduce independently. Event frequency and whether representative event types have been captured affect when sufficient evidence is available. [17]

Observation assignment: watch an approved teaching recording with a supervisor. Make separate rows for video observations, EEG annotations, artifact, and uncertainty. Check synchronization before arguing about which came first. Your objective is to understand the review process, not to diagnose the recording.

Localization and preservation must be considered together.

Neuropsychological and psychological assessment

Standardized assessment establishes cognitive and behavioral baselines, contributes to the clinical formulation, informs risk prediction, and supports counseling and postoperative comparison. It examines more than “which side is weak”: memory, language, attention, executive abilities, psychosocial functioning, and the person’s expectations can all matter. Scores need interpretation in the context of education, language, development, mood, and other influences. Low baseline performance does not mean there is nothing important left to preserve. [18]

Wada testing in selected cases

The Wada procedure temporarily suppresses function in one hemisphere through an arterial medication injection while language and memory are tested. It is invasive and is not required for every patient. Teams decide whether it adds useful information beyond other assessments. Cleveland Clinic describes a total test time of about three to four hours, varying by center. [19]

Electrical stimulation mapping

Specialists can stimulate selected sites and observe effects during tasks such as naming or movement. Mapping may occur during invasive monitoring or an operation. EEG monitoring helps detect stimulation-related discharges or seizures. Duration varies with the territory and tasks; the Epilepsy Foundation describes roughly an hour to several hours. Results inform functional risk rather than guaranteeing that every unmapped site is safe. [21]

SEEG tests a hypothesis. It cannot search the whole brain.

When noninvasive findings are insufficient or conflicting, intracranial recording may clarify a potentially actionable hypothesis. Depth electrodes sample selected three-dimensional locations; subdural grids or strips sample cortical surfaces differently. These are invasive investigations with risks including bleeding, infection, and neurological injury. The possible information gain must justify those risks. [22]

A useful implantation hypothesis includes possible onset sites, competing explanations, and expected spread. Coverage is deliberately selected, not complete. If the true onset lies outside that coverage, the first recorded change can be propagation. This is why implantation planning depends so strongly on the earlier work-up. [23]

Original conceptual diagram. Nodes do not represent actual anatomy, electrode trajectories, or an implantation plan. On a small screen, scroll the diagram horizontally.

After placement, the team verifies electrode identity and location, records events, and may perform functional or seizure stimulation studies. Recording and treatment can be separate admissions. The surgical plan is reconsidered after the findings are integrated; implanting electrodes does not commit the patient to resection. [22]

Look at timing, evolution, anatomy, and the habitual seizure together.

Experts assess intracranial ictal and interictal patterns, their spatial distribution, and their relationship to symptoms. Low-voltage fast activity is one possible seizure-onset pattern; other onset morphologies occur. Neither the largest amplitude nor the highest spike count provides a universal answer. Stimulation findings add context but require their own interpretation. [24]

Low-voltage fast activity can reflect propagation. Symptoms preceding the first recorded change can suggest missed onset. Both observations require reassessing the anatomical and electroclinical hypothesis. [3]

This is a student documentation exercise, not a diagnostic scoring rule.

What about high-frequency oscillations and AI biomarkers?

Promising associations must be distinguished from evidence that biomarker-guided intervention improves outcomes. In the 2022 randomized HFO trial, HFO-guided intraoperative tailoring did not establish non-inferiority to spike-guided tailoring overall. Its setting and results should not be generalized to every HFO method, but it illustrates why a plausible marker requires prospective evaluation. [25]

Onset’s synthetic 120 Hz bursts and interictal features are teaching ingredients. Its contact rankings do not replicate the multimodal clinical decision described here.

Agreement is useful. Contradictions are work to do.

The team reviews the overall evidence and possible benefits and harms before discussing a recommendation with the patient. The available treatment family can include resection, ablation, disconnection, or neuromodulation, depending on the individual condition and goals. Some interventions aim at seizure freedom; others aim mainly to reduce burden. Further evaluation or continued nonsurgical treatment may be the appropriate outcome. [26]

General medical and anesthetic assessment, individualized consent, and recovery planning are also part of preparation. Follow-up looks beyond a single early seizure count: later seizure control, cognition, mood, complications, function, and quality of life matter. Local follow-up schedules and medication decisions vary. [20]

Why this can take months

A scan’s acquisition time is only one part of the pathway. Recording must capture useful events; specialists must interpret and reconcile findings; appointments and multidisciplinary reviews must be coordinated. These clocks should never be added together as if every patient follows the same schedule.

There is no defensible universal “doctor-hours per patient” number in this guide. For an engineering project, measure actual review time, interruptions, repeated annotations, registration checks, and report preparation at the intended service. A recording lasting seven days does not imply that a doctor watches every second at normal speed.

More data do not automatically remove ambiguity

What does “subjective” actually mean?

Experts exercise judgment under incomplete information; that does not make their decisions arbitrary. A prospective study of expert interictal-discharge interpretation found agreement depended on whether the task concerned an individual waveform or an entire EEG. It demonstrates an annotation-reliability problem, not the error rate of epilepsy surgery or a measure of agreement on surgical plans. [29]

Same pathway. Different open questions.

These invented cases are discussion exercises, not decision rules or recommendations for a real patient. Open each case and identify what would change your confidence.

Ten days inside the team’s questions

This is a teaching itinerary spanning different patients and stages of care. It does not imply that one patient completes evaluation in ten days. Observation requires the service’s permission, patient consent where applicable, supervision, and appropriate handling of patient information. Use synthetic or approved de-identified material for assignments.

Save effort without hiding the evidence

The useful engineering question is precise: which task becomes faster or more reliable, for whom, and with what new failure modes? “AI finds the seizure focus” is too broad to evaluate.

Where Onset fits

Onset is a synthetic teaching prototype: signals, features, two rankings, evidence windows, a report and a constrained assistant. It does not ingest real clinical imaging, plan electrode trajectories, model cognitive risk, or establish a safe surgical target. Its useful lesson is how to make a proposed finding inspectable.

An LLM could help retrieve an already verified observation or draft an evidence-linked summary. It must not turn a rank into a surgical instruction. Tool permissions, patient scope, provenance and evaluation matter as much as fluent language. Explore the Qwen teaching lab and model evaluation chapter.

Could you explain the case without overselling it?

1. Explain why referral for surgical evaluation is different from deciding to operate.
2. Distinguish a lesion, interictal-discharge source, recorded SOZ, inferred epileptogenic zone and functional boundary.
3. Explain which question MRI, PET, ictal SPECT, MEG, fMRI and intracranial EEG address.
4. Describe a plausible reason for discordance that does not assume one specialist is wrong.
5. Separate a scan’s acquisition time from event sampling, expert review and pathway waiting.
6. Explain why the first recorded channel is not necessarily the biological origin.
7. Propose one engineering improvement and a study that could show whether it actually helps.

Abbreviations

DRE: drug-resistant epilepsy. EMU: epilepsy monitoring unit. iEEG: intracranial EEG. SEEG: stereoelectroencephalography. SOZ: seizure-onset zone. EZ: epileptogenic zone. IED: interictal epileptiform discharge. FDG: fluorodeoxyglucose. MEG: magnetoencephalography. SISCOM: subtraction ictal SPECT coregistered to MRI. DTI: diffusion tensor imaging. MDT: multidisciplinary team.

Sources you can follow

Guidelines and consensus documents establish the clinical framework; hospital pages provide explicitly local timing examples; primary studies illustrate methodological limits. Sources checked September 19, 2026. This educational synthesis has not undergone independent clinical review.

1. ILAE — Definition of drug-resistant epilepsy (2010)
2. ILAE expert consensus — Timing of referral to evaluate for epilepsy surgery (2022)
3. Schuele et al. — Seminars in epileptology: presurgical epilepsy evaluation (2025)
4. NICE NG217 — Non-pharmacological treatments
5. Baumgartner et al. — Presurgical epilepsy evaluation and epilepsy surgery (2019)
6. Johns Hopkins — Diagnosing seizures and epilepsy
7. Epilepsy Foundation — Tests before epilepsy surgery
8. ILAE Neuroimaging Task Force — Structural MRI recommendations / HARNESS (2019)
9. NICE NG217 — Diagnosis and assessment
10. EANM — Practice guidelines for PET and SPECT in epilepsy (2024)
11. Cleveland Clinic — Magnetoencephalography
12. Primary study — EEG source imaging in presurgical evaluation (2021)
13. American Academy of Neurology — fMRI presurgical evaluation guideline (2017)
14. Winston et al. — Optic-radiation tractography: comparison of methods (2011)
15. Cleveland Clinic — SEEG test
16. Johns Hopkins — Epilepsy monitoring unit
17. Epilepsy Foundation — Video EEG
18. ILAE — Neuropsychological assessment in epilepsy surgery (2019)
19. Cleveland Clinic — Wada test
20. UCLH — A guide for patients considering epilepsy surgery
21. Epilepsy Foundation — Brain mapping
22. Johns Hopkins — Intracranial monitoring for epilepsy
23. Khoo et al. — Technical aspects of SEEG (2020)
24. Frauscher et al. — Learn how to interpret and use intracranial EEG findings (2024)
25. Zweiphenning et al. — HFO trial: randomized comparison of intraoperative guidance (2022)
26. Cleveland Clinic — Epilepsy surgery
27. University Hospitals Birmingham — FDG brain PET patient information
28. Cleveland Clinic — Neuropsychological testing and assessment
29. Jing et al. — Interrater reliability of experts identifying interictal discharges (2020)