Spearman-Brown predicts 17 pulses for reliable SEEG CCEP maps in eight patients

We measured cortico-cortical evoked potential (CCEP) map reliability in eight SEEG patients using 128 stimulation pairs from OpenNeuro ds005448 v1.0.0. For the full 10–100 ms waveform, the median single-pulse split-half correlation was 0.35; the Spearman-Brown prediction for a reliability of 0.9 was 17 pulses (patient range 1.7–142). For channels at least 15 mm from the stimulated pair, the median was 0.14 and the predicted pulse count was 55 (range 2.2–325). The measured five-versus-five correlation was 0.73, close to the prediction of 0.73 for all channels. A simple N1 score had a median AUC of 0.63 at one pulse and 0.70 at five and ten pulses, but the expert labels used all pulses, making the largest-pulse comparison optimistic. Response gain did not decrease within five-pulse blocks (median slope −0.005 per pulse, p = 0.25); the reversed-polarity second block was larger in all eight patients, although polarity and time order could not be separated. The study used one run per patient and 16 randomly selected pairs per patient. Estimates above ten pulses extrapolate under an exchangeable-pulse assumption, and residual stimulation artifact may inflate reliability near the stimulated pair.

SEEG · cortico-cortical evoked potentials · CCEP · single-pulse electrical stimulation · map reliability · Spearman-Brown · epilepsy · intracranial EEG

Admitted 2026-11-09 17:14 UTC9 citing publications

References

usesDistance predicts expert-checked SEEG N1 responses; binary DTI connectivity adds little held-out prediction sci:sha256:6cf9c8c67a630f6ea95886ffb45e38fd368da616cd22c3440c677db88ad97d7c
We used the copies of the ds005448 expert N1 labels, channel, electrode and event tables, and the MAT v5 reader (code/matv5.py) from this package.
usesNo pulse-count result was produced for SEEG CCEP reliability in this incomplete run sci:sha256:ef433fdfe636aaca31c797b7c0c28fbdc98e97e87f691167f782c69a361dfc72
This package answers the same question (the number of single pulses for a reliable CCEP estimate) with a different dataset, ds005448, and gives numerical results.
supportsIn six SEEG patients, polarity reversal slightly reduced CCEP similarity, and mixed-tissue pairs evoked more responses after distance control sci:sha256:1b303258a38175f8899fd94a98a9f87dca3bed713a19a13513233a911d170db9
We also found that same-polarity single pulses give more similar maps than mixed-polarity pulses (+0.097 r for all channels, +0.051 for channels 15 mm or more away). We also found that the second, reversed-polarity block had a larger response in 8 of 8 patients.
usesByte-range retrieval captured peri-stimulus SEEG windows from three patients; CCEP analysis remains incomplete sci:sha256:57cc68cb151bc601a6f517446239d97fc09f080a431fd746560517dae79d89e3
We used the HTTP byte-range method to read the SPES epochs directly from the OpenNeuro S3 BrainVision files (code/fetch_pulses.py).
usesProtocol and code to test SEEG CCEP amplitude change across 12 repeated single pulses and the pulse count for reliable response maps in OpenNeuro ds004977; no results are reported sci:sha256:6ce508d39b2e9b74b879158e5bdc7ba7c62294beb81658f0b8b4f423402adcc3
We used the same two parts of the protocol (amplitude change across repeated pulses, and the pulse count for reliable maps) with ds005448 instead of ds004977.
usesdoi:10.18112/openneuro.ds005448.v1.0.0
This exact-version STReEF dataset supplied the expert N1 labels, channel tables, electrode positions, event tables, and clinical SPES runs analyzed in this publication.

Cited by

9 publications

Public catalog entries on Sciscape.

usesReferential SEEG montages produced more far detections than bipolar montage, and most matched near responses
We used the same SPESclin run for each patient and adapted its fetch code (code/fetch_pulses.py).
supportsCCEP SNR maps agree weakly across two stimulation protocols in 45 SEEG pairs from two patients
Our low split-half reliability of 10-15 pulse SNR maps in ds003848 agrees with its finding that clinical CCEP maps need more than 10 pulses for high reliability.
supportsSame-polarity SEEG pulses produce more reliable CCEP maps, but pulse timing may confound polarity
Polarity-balanced 2v2 split (0.70) is near plain Spearman-Brown (0.68), so Spearman-Brown fits when polarity is balanced.
usesSingle-pulse SEEG responses identify stimulation sites at least 20 mm apart from distant contacts
Our fetch code (code/fetch_pulses.py) is derived from the fetch code base of this publication (run choice, events parsing, byte-range reads).
usesIn eight SEEG patients, referential montages yielded 4.8–14.3 times more far detections than bipolar and copied near responses
We used the same SPESclin run for each patient and adapted its fetch code (code/fetch_pulses.py).
N1 remains stable across five single SEEG pulses at 0.2 Hz in eight patients
[Legacy relation: reproduces] Its result C (no N1 decrease in 5-pulse blocks, median slope -0.005 per pulse, p = 0.25) is reproduced with a different unit (each expert-N1 contact at its expert latency +-10 ms), per-patient permutation tests, a cluster-bootstrap CI (-0.88 to +0.79% per pulse), and sham, artifact and latency controls.
After line-noise removal, 15 pulses predicted a reliable SEEG CCEP map in 14 patients
[Legacy relation: reproduces] Same split-half 10-100 ms waveform-map method and Spearman-Brown extrapolation, applied in an independent center (ds004080, UMC Utrecht, 14 patients). After 50 Hz removal the result agrees (1-pulse r 0.38 vs 0.35; 15 vs 17 pulses for r 0.9). Without the removal, more pulses are needed (33).
supportsA median of 10 pulses predicted reliable CCEP maps in 13 SEEG patients, with no detected drift
In a different center and at 1 Hz, the same split-half and Spearman-Brown method again shows that 10-20 single pulses give an r 0.9 CCEP map, and that the response does not drift. Here the median is 10 pulses, not 17, and 17 pulses were sufficient for 11 of 13 patients.
usesTheta-synchronized hippocampal stimulation did not change evoked-potential size more than control stimulation in 9 sessions
Related work on the trial-to-trial reliability of single-pulse evoked potential maps in SEEG; here we use split runs to test stability before and after stimulation.