Interictal Epileptiform Discharges

The interictal epileptiform discharge (IED) is the single most consequential pattern in clinical electroencephalography. A genuine IED, captured between seizures in a patient who is behaviorally and electrographically normal, is among the most specific noninvasive biomarkers of an epileptogenic cortical network. Yet the same morphology is mimicked by a long list of benign variants, physiologic transients, and artifacts, and overcalling a discharge can label a patient with epilepsy for life, restrict driving and employment, and commit them to years of unnecessary medication. The discipline of IED interpretation therefore rests on a precise, reproducible vocabulary of morphology, field, and context, and on the humility to recognize that the term epileptiform is a probabilistic statement, not a binary one. This module builds that vocabulary from the membrane upward, connects each waveform feature to the cortical generator that produces it, and closes with the quantitative and computational tools that are reshaping spike interpretation in the mid-2020s.

It is worth stating at the outset what an IED is and is not. It is electrographic evidence that a population of cortical neurons can, between seizures, fire in pathological synchrony. It is not a seizure, it is not proof that the patient has had a seizure, and its absence is not proof that the patient is free of epilepsy. The relationship between the interictal spike and the clinical disease is statistical and indirect, mediated by the size of the irritative zone, the orientation of its generating cortex relative to the skull, the patient's age, the state of vigilance during recording, and the prior probability established by the clinical history. Holding all of these in mind simultaneously is what separates the expert reader from the pattern-matcher, and it is the through-line of everything that follows.

Morphologic criteria: what makes a discharge epileptiform

The International Federation of Clinical Neurophysiology defines an epileptiform discharge by a constellation of features, no single one of which is sufficient. The waveform must be paroxysmal, meaning it stands out abruptly from the ongoing background rather than emerging gradually. It must have a pointed or spiky peak and an asymmetric, often biphasic or triphasic contour in which the initial sharp deflection differs in slope from its recovery limb. Duration is the formal classifier: by the IFCN convention a spike measures roughly 20 to less than 70 milliseconds, whereas a sharp wave spans 70 to 200 milliseconds. This is a distinction of timing alone and carries no difference in epileptogenic meaning; both denote the same underlying process recorded with slightly different temporal dispersion. The duration is measured at the base of the transient, not at half-amplitude, and the boundary is a convention rather than a biological discontinuity, so a transient at the 70 millisecond border should be judged by its company rather than by the stopwatch alone.

Beyond the transient itself, three contextual features carry enormous weight. First, a true discharge possesses a physiologic field: the potential must be visible across more than one electrode in a spatially coherent gradient, with a clear point of maximal negativity that decays smoothly to neighboring contacts. A deflection confined to a single electrode is almost always artifact, because no plausible cortical dipole projects to one contact alone while leaving its immediate neighbors untouched. Second, the discharge typically has surface negativity at its peak, reflecting the orientation of the underlying current source, with the apical dendrites of pyramidal cells acting as the negative pole at the cortical surface. Third, a convincing IED is usually followed by an after-going slow wave, a lower-frequency surface-negative deflection that represents the inhibitory aftermath of the synchronized excitation. Finally, the discharge should disrupt the background, interrupting the ongoing rhythm rather than riding seamlessly on top of it the way a benign rhythmic variant does.

A subtler but increasingly emphasized criterion is the di-phasic or tri-phasic field with a logical dipole. When a reader inspects a candidate spike across multiple montages, the negativity should reverse phase across a consistent electrode in bipolar chains and should show a coherent maximum in referential recording, and that maximum should correspond to a plausible cortical generator. A transient whose apparent field jumps illogically between non-adjacent electrodes, or whose polarity cannot be reconciled across montages, is far more likely to be artifact or a benign variant than a true discharge. This is why the seasoned reader never adjudicates a spike on a single montage and never trusts the automatic detector without confirming the field by eye.

Inter-rater agreement studies are sobering and clarifying in equal measure. When expert electroencephalographers independently score candidate transients, agreement on whether a given waveform is truly epileptiform is only moderate, with kappa values that rise substantially when the six-feature criteria are applied explicitly and when readers are forced to commit to a confidence level. This empirical fact has a direct clinical consequence: a report that calls a transient epileptiform without describing the features that justify the call is nearly impossible for a downstream clinician to weigh. The modern standard, embodied in structured reporting templates, is to name the discharge, state its field and morphology, and where appropriate signal the reader's confidence, so that the treating physician can integrate it into a probabilistic rather than a categorical decision.

The paroxysmal depolarizing shift: from neuron to scalp

The cellular substrate of the interictal spike is the paroxysmal depolarizing shift (PDS), a stereotyped intracellular event recorded from epileptogenic neurons. A PDS is a large, sustained depolarization on the order of 20 to 50 millivolts lasting roughly 50 to 200 milliseconds, crowned by a high-frequency burst of action potentials and terminated by a powerful after-hyperpolarization. The depolarizing envelope is now understood as a giant excitatory postsynaptic potential driven by recurrent glutamatergic circuitry and amplified by voltage-gated inward currents and NMDA-receptor-dependent conductances; the terminating after-hyperpolarization is generated by calcium-activated and other potassium conductances together with GABA-B-mediated inhibition. The after-hyperpolarization is the cellular correlate of the scalp after-going slow wave, and the progressive failure of this inhibitory brake is one of the conceptual bridges between the interictal spike and the transition to seizure.

What converts an isolated PDS in a few neurons into a synchronized population event is the failure of surround inhibition. In healthy cortex, a burst of pyramidal activity recruits feedforward and feedback GABAergic interneurons that clamp down on neighboring neurons and prevent the burst from spreading, an inhibitory restraint that confines excitation in both space and time. In epileptogenic cortex this restraint is impaired by a combination of interneuron loss or dysfunction, altered chloride homeostasis that can render GABA-A signaling less hyperpolarizing or even depolarizing, abnormal gap-junction coupling, and pathological recurrent excitatory connections such as the aberrant mossy fiber sprouting seen in mesial temporal sclerosis. When surround inhibition fails, the PDS recruits its neighbors rather than being contained by them, and a population of neurons fires in the near-simultaneous burst that the scalp records as a spike. The interictal discharge is thus a snapshot of a microcircuit in which excitation has transiently outrun inhibition but inhibition has, for now, won the race to terminate the event.

A PDS in a handful of neurons produces no scalp signal whatsoever. The transition from a cellular event to a scalp-recordable IED requires synchronization across a critical mass of cortex. Classic estimates place the lower limit near 6 square centimeters of synchronously active gyral cortex, but simultaneous intracranial-to-scalp correlation studies show that reliable scalp visibility generally requires considerably more, on the order of 10 square centimeters or greater, with synchronous activation of roughly 10 to 20 square centimeters being common for spikes that surface; in one well-known series about 90 percent of cortical spikes with a source area exceeding 10 square centimeters produced a scalp correlate while only about 10 percent of smaller-area spikes did. Smaller generators, or those confined to sulcal walls whose dipoles are tangential to the skull, may produce robust intracranial spikes that are entirely invisible on routine EEG. This spatial filtering is the deepest reason that scalp EEG underestimates the true irritative zone and that a normal study never excludes epilepsy. The corollary, demonstrated repeatedly in simultaneous scalp and intracranial recordings, is that only a minority of intracranially recorded spikes have a scalp counterpart, and the ones that do are biased toward large, superficial, radially oriented generators.

Discharge types: spike-and-wave, polyspikes, and their meaning

When spikes recur in a rhythmic, repetitive train coupled to slow waves, the pattern becomes a spike-and-wave complex. The classic 3 Hz generalized spike-and-wave of childhood absence epilepsy is the prototype: bilaterally synchronous, frontally predominant, with an abrupt onset and offset and a frequency that typically begins near 3 to 4 Hz and decelerates toward 2.5 to 3 Hz over the run. This pattern is generated by the thalamocortical loop, in which the rhythmic interplay of cortical pyramidal cells, thalamic relay neurons, and the GABAergic thalamic reticular nucleus, gated by low-threshold T-type calcium currents, imposes a synchronous oscillation across both hemispheres. The same machinery that produces the normal sleep spindle is, in this model, hijacked and run at a pathological gain, which is why genetic generalized epilepsy can arise from an intact brain rather than a structural lesion. Slow spike-and-wave at 1.5 to 2.5 Hz carries a very different prognosis, being a signature of Lennox-Gastaut syndrome and a marker of diffuse encephalopathy.

Polyspikes, or polyspike-and-wave complexes, consist of two or more spikes clustered before the slow wave and are strongly associated with the idiopathic (genetic) generalized epilepsies, particularly juvenile myoclonic epilepsy, where they correlate with the myoclonic phenotype and are accentuated by sleep deprivation, the early-morning state, and photic stimulation. The number, regularity, and reactivity of polyspikes therefore carry syndromic information that a single morphologic glance can capture, and the count of spikes per complex has even been studied as a rough correlate of the likelihood of an accompanying myoclonic jerk. Focal discharges, by contrast, derive their meaning chiefly from where they project rather than their internal complexity, although focal polyspikes can mark particularly active or structurally abnormal cortex, as in the ribbon-like discharges seen over a focal cortical dysplasia.

Two named focal patterns deserve specific mention because of their localizing power. Temporal intermittent rhythmic delta activity (TIRDA) is a run of rhythmic delta over the temporal region that, despite not being a spike, carries an epileptiform connotation nearly as strong as a temporal sharp wave and is a recognized interictal marker of temporal lobe epilepsy. At the catastrophic end of the spectrum, periodic lateralized epileptiform discharges, now standardized as lateralized periodic discharges (LPDs) in the ACNS nomenclature, mark an acutely irritable hemisphere, classically in herpes simplex encephalitis, acute stroke, or other acute structural insults, and sit on the ictal-interictal continuum discussed in a later module. Recognizing these named patterns expands the reader's vocabulary beyond the simple spike and sharp wave and is part of the modern standardized approach.

Localization, yield, and the irritative zone

For focal discharges, the topographic field is the engine of localization. The electrode of maximal surface negativity, the steepness of the gradient to surrounding contacts, and the polarity reversal across a bipolar montage together triangulate the irritative zone, the cortical territory capable of generating interictal spikes. Crucially, the irritative zone is one of several distinct cortical zones in the presurgical framework and is not identical to the seizure-onset zone, the epileptogenic zone whose removal abolishes seizures, the symptomatogenic zone that produces the clinical signs, or the functional deficit zone. The irritative zone is typically the largest of these and may be displaced from the region whose resection actually renders the patient seizure-free, which is precisely why interictal spikes alone are insufficient for surgical decision-making and must be integrated with ictal recordings and imaging. An anterior temporal sharp wave maximal at F7 or T1, for instance, strongly implicates mesial temporal structures even though the recording contact overlies neocortex, because the deep mesial dipole projects to the inferolateral scalp. Use the interactive localization model below to see how a single dipole orientation dictates the scalp field and the bipolar phase reversal that a reader uses to name the source.

The diagnostic yield of a single routine EEG in capturing IEDs in a patient with established epilepsy is only on the order of 30 to 50 percent; serial studies, sleep deprivation, and prolonged or ambulatory recording progressively raise sensitivity toward the 80 to 90 percent range because activation procedures and natural sleep dramatically increase spike rates. Sleep is the single most powerful activator, since the synchronizing influence of non-REM sleep on thalamocortical circuits lowers the threshold for spike generation, which is why a sleep-deprived study that captures drowsiness and stage N2 sleep is so much more sensitive than a brief awake recording. Hyperventilation and photic stimulation add yield in specific syndromes, particularly the genetic generalized epilepsies. The practical lesson is that a normal first EEG should never close the question in a patient with a convincing clinical history; the appropriate next step is a longer or sleep-deprived study, not a conclusion that the patient does not have epilepsy.

The complementary metric is specificity. In healthy adults without epilepsy, the prevalence of incidental epileptiform discharges is low, on the order of a fraction of a percent, but it is meaningfully higher in children, in relatives of patients with genetic generalized epilepsy, and in selected populations such as those with prior brain injury or certain psychiatric conditions. This age and population dependence is not a footnote; it is central to interpretation, because the same waveform means something quite different in a healthy 40-year-old than in an 8-year-old, in whom benign discharges are more common and the centrotemporal spike of self-limited epilepsy may appear in children who never seize.

The Bayesian framing also disciplines the use of the EEG after a first seizure. Guideline-level evidence establishes that interictal epileptiform discharges on the EEG roughly double the risk of seizure recurrence after a first unprovoked event, which is one of the factors that can shift management toward starting an antiseizure medication earlier rather than waiting for a second seizure. The EEG thus functions not merely as a diagnostic test but as a prognostic one, and the reader who understands this will weigh the decision to call a borderline transient epileptiform with full awareness that the call may change whether the patient is treated.

Quantitative markers and computational reading in 2026

The visual read remains the gold standard, but it is increasingly augmented by quantitative tools. The most clinically important development of the last decade is the recognition of high-frequency oscillations (HFOs), brief bursts of activity in the ripple (roughly 80 to 250 Hz) and fast ripple (roughly 250 to 500 Hz) bands. Fast ripples in particular appear to be more specific markers of the epileptogenic zone than conventional spikes, because they reflect the pathologically synchronous firing of small neuronal clusters and tend to cluster more tightly around the seizure-onset zone. HFOs are largely an intracranial phenomenon and a research and presurgical tool rather than a routine scalp measurement, since recording frequencies up to several hundred hertz reliably requires depth or grid electrodes and high sampling rates, and scalp detection is confounded by muscle artifact in the same band. Their importance for the trainee is conceptual: they demonstrate that the conventional spike is only one window onto epileptogenic tissue and that information above the traditional EEG bandwidth carries localizing value.

Electrical source imaging (ESI) has matured into a clinically validated adjunct. By recording with high-density arrays of 64 to 256 electrodes and solving the inverse problem against an individualized head model derived from the patient's own MRI, ESI estimates the cortical source of averaged interictal spikes with an accuracy that, in well-conducted studies, contributes meaningfully to presurgical localization and can be concordant with the resected zone in patients who become seizure-free. ESI does not overturn the field logic taught in this course; it formalizes and refines it, replacing the reader's mental triangulation with a quantitative distributed-source estimate while remaining bound by the same physics of volume conduction and the same ill-posedness of the inverse problem.

Automated spike detection powered by deep learning has reached the point of clinical deployment as a triage and second-reader tool. Convolutional and transformer-based detectors trained on large expert-annotated archives can flag candidate discharges across hours of recording, reducing review time and improving sensitivity for sparse spikes, and several systems have been cleared for clinical use as aids rather than replacements. The controversies are instructive and are very much live in 2026: detectors trained on one population or one acquisition system can generalize poorly to another, they inherit the moderate inter-rater disagreement of their training labels, and an over-trusted automatic call can launder a benign variant into a diagnosis of epilepsy. The expert consensus is that these tools augment but do not replace the human reader, who must confirm every machine-flagged transient against the six-feature gestalt and the clinical context. The genuinely hard, unsolved problem is not detecting a clear spike but adjudicating the borderline transient, and that is exactly where both humans and machines remain fallible.