Page 1890 · Картування зони початку нападу за ЕЕГ — spectral, morphological, and propagation analysis of ictal discharge to localize the seizure onset zone
Every downstream analysis in ictal onset mapping — pattern classification, spectral decomposition, propagation tracking — depends entirely on one upstream requirement: synchronous, high-fidelity, multi-channel raw EEG. Whether the array is 19 scalp electrodes or 150+ stereo-EEG (SEEG) depth contacts, every channel must be sampled against the same clock, referenced consistently, and stored losslessly, because the millisecond-scale timing differences between channels are themselves the signal that reveals where a seizure begins and where it goes next.
Two acquisition paradigms feed onset zone mapping:
Scalp video-EEG (non-invasive, screening): • 10-20 or extended 10-10 system: 19–32 electrodes • Referential montage during acquisition (common reference or average reference), re-montaged offline to bipolar chains for review • Limited spatial resolution: scalp electrode "sees" a smoothed, volume-conducted mixture of activity from several cm² of underlying cortex • Onset patterns are often poorly localizing at scalp level — a true focal fast discharge can appear only as diffuse attenuation ("electrodecrement") due to spatial low-pass filtering by skull and scalp
Intracranial EEG — SEEG or subdural grids/strips (invasive, pre-surgical): • SEEG: stereotactically implanted depth electrodes, each with 8–18 contacts along its length, sampling both cortical surface and deep structures (insula, cingulate, hippocampus) • ECoG grids/strips: subdural electrodes over cortical surface, denser 2D sampling of gyral activity but limited to accessible convexity • Bipolar referencing (adjacent-contact subtraction) removes common noise and far-field volume conduction, sharpening the spatial specificity needed for onset-channel identification • Sampling at 1–2kHz (vs 256Hz scalp) is essential because ictal high-frequency oscillations extend to 250–500Hz — Nyquist requires ≥2× that rate to avoid aliasing
Signal chain requirements: • Anti-aliasing hardware filter before ADC (low-pass at ~40% of sampling rate) • 16–24 bit ADC resolution: dynamic range must capture both µV-scale background rhythms and mV-scale artifact without clipping • Common clock distribution across all amplifier boards: SEEG systems with >100 channels split across multiple headboxes must be hardware-synchronized, or propagation-lag estimates between channels on different boards become meaningless • Continuous, gap-free recording across the entire monitoring admission (days), since the analyst does not know in advance which seizure will be most informative for onset mapping
The very first few seconds of electrographic change at seizure onset have a shape, and that shape is not random. Decades of intracranial EEG-pathology correlation studies have shown that distinct onset morphologies — low-voltage fast activity, rhythmic spike-and-wave discharges, sharp-and-slow-wave complexes, and several rarer patterns — associate with different underlying substrates and carry different implications for how confidently the onset channels can be trusted as the true epileptogenic zone.
Low-voltage fast activity (LVFA): • Abrupt attenuation of background rhythms ("electrodecrement") followed immediately by low-amplitude, high-frequency (>13Hz, often 20–40Hz) rhythmic discharge • Strongly associated with focal cortical dysplasia (FCD type II) and other malformations of cortical development • Considered the single most reliable electrographic marker of a discrete, resectable epileptogenic zone — surgical series consistently show higher seizure-freedom rates when LVFA is the recorded onset pattern • Reflects a highly synchronized, spatially compact population of hyperexcitable neurons recruiting rapidly
Rhythmic spike-and-wave / rhythmic sharp activity: • Regular, repetitive spike (or sharp wave) followed by a slow wave, at 2–5Hz, gradually increasing in frequency and decreasing in amplitude ("evolving" discharge) • Classic pattern of mesial temporal lobe epilepsy arising from the sclerotic hippocampus • Localizing value is good but less absolute than LVFA — the onset zone often includes a broader mesial temporal network (hippocampus, entorhinal cortex, amygdala) rather than a single discrete focus
Sharp-and-slow-wave complexes / bursts of polyspikes: • Higher amplitude, lower frequency (1–4Hz) discharges, often less rhythmically regular than the spike-wave pattern • Associated with more diffuse or multifocal networks, or with rapid secondary propagation obscuring the true focal origin • Least reliable single-pattern predictor of a resectable zone; typically prompts more weight on the spectral and propagation evidence (Stages 3–4) before committing to a surgical target
Classification workflow in practice: • Epileptologists mark the earliest channel(s) showing unequivocal, sustained rhythmic change distinct from background • Pattern is scored qualitatively (visual review) and increasingly with automated morphological classifiers (amplitude, dominant frequency, rhythmicity index, spectral edge frequency) trained on expert-labeled onset segments • A single patient may show different onset patterns across different seizures — multi-seizure concordance strengthens localization confidence considerably more than any single seizure's pattern alone
Visual inspection of raw voltage traces captures only part of the story. Converting each channel into a time-frequency spectrogram — via short-time Fourier transform or continuous wavelet transform — exposes energy concentrated in frequency bands invisible or ambiguous to the naked eye. The most clinically significant discovery in this domain has been the ictal high-frequency oscillation (HFO): brief bursts of 80–500Hz activity that frequently precede or coincide with visible seizure onset, and that cluster with striking spatial precision over the true epileptogenic tissue.
Time-frequency decomposition methods:
1. Short-time Fourier transform (STFT): • Signal divided into overlapping windows (e.g., 1s, 90% overlap); FFT computed per window • Produces a 2D image: time × frequency × power (dB), i.e. the spectrogram • Fixed time-frequency resolution tradeoff (Heisenberg-like uncertainty): wider windows resolve frequency better but blur timing of rapid onset transitions
2. Continuous wavelet transform (CWT, e.g. Morlet wavelets): • Adaptive resolution: better time resolution at high frequencies, better frequency resolution at low frequencies • Preferred for capturing brief, transient HFO bursts (tens of milliseconds) that STFT with a fixed window can smear or miss
3. Matching-pursuit / multitaper methods: • Used in dedicated HFO-detection pipelines for higher spectral purity and reduced leakage from adjacent bands
Interpreting the ictal spectrogram: • Background interictal spectrum: broadband, dominated by lower frequencies (delta–alpha), relatively flat power beyond ~30Hz • At true onset channels: a vertical band of elevated power appears simultaneously across a wide frequency range — from beta/gamma up through ripple (80–250Hz) and sometimes fast-ripple (250–500Hz) — often several hundred milliseconds to seconds before the discharge is visible as a clear rhythmic pattern in the raw trace • This spectral "flare" is the electrophysiological correlate of a hypersynchronous neuronal population recruiting — the same population later expressed visually as LVFA or spike-wave morphology • Automated HFO detectors (e.g., energy-threshold, line-length, or short-line-length algorithms) flag discrete ripple/fast-ripple events and compute an HFO rate per channel; channels with the highest ictal HFO rate co-localize with resection margins in the majority of seizure-free surgical outcomes
Caveats: • Muscle and electrode artifact can mimic high-frequency power — spectral analysis must be paired with morphological review (Stage 2) to exclude non-cerebral contamination • Physiological ripples (80–200Hz) occur normally in some structures (e.g., hippocampus); pathological significance requires ictal timing and rate elevation above interictal baseline, not frequency alone
A seizure is rarely confined to its onset channels for long. Within one to several seconds, the ictal discharge recruits neighboring cortex, then more distant, synaptically connected regions — spreading along preferential anatomical pathways rather than uniformly in all directions. Reconstructing this spread, channel by channel and millisecond by millisecond, both confirms which channels were first (the onset zone) and reveals the functional network the epileptogenic tissue is embedded in — critical information for predicting seizure semiology and surgical risk.
Quantifying spread from the onset channels:
1. Pairwise cross-correlation with lag search: • For every channel pair (i, onset-channel), compute the cross-correlation of band-limited amplitude envelopes (or ictal power) across a range of time lags • The lag maximizing correlation is taken as the propagation delay from the onset channel to channel i • A monotonically increasing lag with anatomical/network distance from onset supports a genuine propagating wavefront rather than simultaneous independent activation
2. Onset-time mapping (leading-edge detection): • Per-channel "involvement time" defined as when a chosen ictal marker (rhythm amplitude, HFO rate, or line-length statistic) crosses a threshold above its own baseline • Sorting all channels by involvement time produces a spatiotemporal recruitment order — visualized as a color-coded map (earliest = onset zone, progressively later shades tracing the spread)
3. Directed connectivity measures: • Phase-lag index / phase-slope index: estimate directionality of coupling between channel pairs while being robust to volume-conduction (zero-lag) artifacts • Granger-causality-style autoregressive models: test whether past activity in channel A improves prediction of channel B beyond B's own past — directional evidence for A→B propagation • These measures build a directed graph over the electrode array: nodes = contacts, edges = propagation direction and strength
Interpreting propagation topology: • Fast, near-simultaneous recruitment of many channels (short lags, low propagation-speed slider values) suggests a highly excitable, densely interconnected network — often lowering confidence that the very first channels represent a small, resectable focus, since the "true" origin may be embedded within a larger rapidly-synchronizing region • Slow, stepwise recruitment with clear, increasing lags to a limited number of channels supports a compact, well-demarcated onset zone with a discrete propagation pathway — the pattern most favorable for surgical planning • Propagation pathways recurring consistently across multiple seizures in the same patient are weighted far more heavily than a pathway seen once, since single-seizure networks can be influenced by state, medication taper, or stimulation artifact from the recording itself
No single feature — not morphology, not spectral signature, not propagation topology — is individually sufficient to declare a resection target. The final stage of ictal onset zone mapping combines all three lines of evidence into a composite, confidence-scored map overlaid on the patient's electrode anatomy, translating raw signal analysis into the single most consequential decision of the pre-surgical workup: which tissue, if resected or ablated, is most likely to render the patient seizure-free.
Composite confidence scoring:
• Pattern score: weighted by onset morphology — LVFA scores highest (most literature support for focal, resectable substrate), rhythmic spike-wave intermediate, sharp-and-slow-wave lowest • Spectral score: weighted by ictal HFO rate and spectral concentration at the presumed onset channels relative to surrounding tissue — a sharp, spatially confined HFO "hotspot" raises confidence; diffuse or absent HFO elevation lowers it • Propagation score: weighted by how compact and temporally well-ordered the recruitment map is — few channels, clean increasing lags = high score; many channels recruited near-simultaneously = low score • The three sub-scores are combined (commonly a weighted sum or logistic combination validated against surgical outcome data) into a single per-channel confidence percentage, then rendered as a spatial heat-map with confidence contours over the electrode implantation scheme or head model
Multi-seizure concordance: • Because any single seizure's onset can be influenced by state (sleep/wake), anti-seizure medication levels during monitoring, or stimulation-induced seizures, epileptologists require concordance across multiple spontaneous, habitual seizures before finalizing the onset zone • Concordant onset location across ≥3 seizures substantially increases confidence and is a standard reporting requirement in most epilepsy surgery conferences • Discordant onset zones across seizures (a "multifocal" picture) sharply reduce surgical candidacy and often redirect the plan toward palliative neuromodulation (responsive neurostimulation, vagus nerve stimulation) rather than resection
From map to decision: • High-confidence, tightly-clustered, LVFA-pattern onset zones anatomically consistent with MRI-visible lesions (e.g., FCD) are the strongest candidates for curative resection or laser interstitial thermal therapy • Moderate-confidence network-level zones (e.g., mesial temporal) support standard resections (anterior temporal lobectomy, selective amygdalohippocampectomy) with well-established outcome literature even without single-channel-level precision • Low-confidence, diffuse, or discordant maps are the cases most likely to proceed to chronic responsive neurostimulation, where the device itself continuously re-analyzes onset patterns in the same conceptual pipeline (pattern + spectral trigger detection) to deliver closed-loop therapy rather than a one-time resection
The core clinical logic: localization confidence is never asserted from a single feature. A low-voltage fast onset pattern with a sharp, spatially-confined high-frequency oscillation burst and a slow, compact propagation pathway — concordant across multiple seizures — is the electrographic signature most consistently associated with post-surgical seizure freedom in modern epilepsy center outcome series. Every stage of this pipeline exists to either strengthen or undercut that composite confidence before a single millimeter of brain tissue is proposed for resection.