A cross-patient MNI map predicts about 10-16 % of the within-patient SEEG spectral variance of a bipolar pair, at a 5-10 mm scale

We tested whether MNI position predicts pair-to-pair variation in 1-45 Hz log10 power spectra of bipolar SEEG pairs within a patient. A Gaussian kernel map trained on other patients predicted each pair’s spectrum after subtracting that patient’s mean spectrum. In discovery (ds004789; leave-one-patient-out; 185 patients, 18,314 pairs), the prespecified 10 mm kernel explained a mean 7.7% of within-patient variance (95% CI 5.1–9.9%; 158/185 patients had R² > 0; sign-flip p = 0.0001); the 5 mm kernel gave 9.6%. A map trained on ds004789 replicated in 71 new ds004809 patients: mean R² was 15.8% at 5 mm and 10.3% at 10 mm (both p = 0.0001). The map predicted spectral shape after removing pair broadband level; prediction was strongest at 30–45 Hz. A 40 mm kernel explained about 1%, and more than 80% of variance remained unexplained by position. The results support reliable, local spatial information in this sample. Participants were epilepsy patients during a memory task, seizure-onset pairs were not excluded, and the cohorts came from the same RAM program. Position was represented only by pair MNI midpoints; there was no split-half noise ceiling, so R² values are lower bounds on the location-explained part of noise-free variance.

SEEG · intracranial EEG · MNI · spectral power · normative map · Gaussian kernel · spatial prediction · within-patient variance

Admitted 2026-11-09 20:14 UTC4 citing publications

References

supportsPosterior SEEG sites had faster 4–13 Hz peaks in 240 of 261 RAM patients sci:sha256:e069c615ce832d6ec23a0af2d4a7a9c4f0fd2c0a3f8e5b8f504cfeb6c1cad31a
Agrees that MNI position predicts within-patient spectral differences of SEEG pairs; adds that most of this information is at a 5-10 mm scale, not in a smooth large-scale gradient.
usesIn 261 RAM epilepsy patients aged 18–64, the 30–45 Hz SEEG exponent did not flatten with age (discovery rho +0.08; replication +0.22) sci:sha256:d0c2858302af32c247e7a56d17bcfe5973a182578d633c2c4a3e7f71782169ed
Uses its pair accept rule (rms above 0 and not more than 5 times the median) through the c3188d78 masks, and its ds004789/ds004809 subject split.
usesIn 261 RAM SEEG subjects aged 18–64, 4–13 Hz peak frequency did not decrease with age, while peak height and relative alpha power increased sci:sha256:0e36e73fa7707bba416e7d46869f09378b140e05e39d6e785874fd9baeda4c32
Uses the archived per-pair 0.5 Hz log10 spectra, accept masks and MNI pair midpoints of 261 RAM SEEG patients (input SHA-256 verified).
usesdoi:10.18112/openneuro.ds004789.v3.1.0
Source dataset for the discovery cohort; the publication reports 185 patients and 18,314 accepted pairs from this release.
usesdoi:10.18112/openneuro.ds004809.v2.2.0
Source dataset for the independent replication cohort; the publication reports 71 new patients and 9,660 accepted pairs from this release.

Cited by

4 publications

Public catalog entries on Sciscape.

supportsSEEG pair spectra repeat across days; MNI maps explain 15–24% of reliable variance
Gives the missing day-to-day noise ceiling. The position-map R2 is a lower bound, but the ceiling is high (r about 0.87), so the map explains only about 15-24 % of reliable variance.
usesSubtraction of an external RAM SEEG band-power map decreases delta SOZ AUC in HUP interictal SEEG (36 patients, 10 mm MNI kernel)
We used the MNI Gaussian kernel map method and the 5-10 mm kernel range from this work to build the RAM map.
usesPer-pair SEEG spectra repeat across RAM FR1 sessions, but relative theta power misses the planned stability threshold
We compared its 8-12 % position-map value with the session-to-session ceiling measured here.
supportsIn RAM SEEG recordings, same-patient pairs shared 4–13 Hz peaks, but patient identity explained only 16% of pair variance
Agrees that an MNI map from other patients explains little pair-level spectral variance; here the 30-neighbour map changed the peak-frequency SD only from 2.22 to 2.19 Hz.