Four pulses captured a median 90% of the above-chance AUC of a 10-pulse SEEG N1 map in 8 patients
We evaluated how many single-pulse electrical stimulation (SPES) pulses are needed to reproduce expert N1 labels in SEEG cortico-cortical evoked potential maps. In 8 patients from OpenNeuro ds005448 v1.0.0, we analyzed 157 stimulated pairs and 13,794 pair-contact tests. A negative-peak z-score from the average of 1–10 pulses was compared with expert labels using ROC AUC; group tests were exact two-sided Wilcoxon tests over patients. Median AUC was 0.711 for 1 pulse, 0.773 for 4 pulses, and 0.808 for 10. Four pulses (range 2–7) reached a median 90% of the above-chance AUC of 10 pulses; 7 pulses (range 3–9) reached 95%. Each pulse count below 10 had lower AUC than 10 pulses, with 8/8 patients lower for counts up to 6. Pulse polarity and order did not change detection (median AUC differences within ±0.01; p > 0.4). For polarity-balanced halves, split-half waveform map correlation was 0.719 with 2 pulses per half and 0.820 with 4, below the Spearman–Brown prediction of 0.836 in 8/8 patients (p = 0.0078). Single-pulse N1 amplitude did not decrease within five-pulse polarity blocks (p = 0.84); the first block was larger in 7/8 patients, but polarity and block order were confounded. Interpretation is limited by expert labels based on the average of all pulses, which gives k = 10 agreement a positive bias; the detector scores negative peaks although only 21–65% of expert N1 amplitudes were negative. The study used one center, one current per pair, and 1–2 mA monophasic pulses. These results describe agreement with the full-protocol expert map, not recovery of true connectivity.
SEEG · cortico-cortical evoked potentials · single-pulse electrical stimulation · N1 · effective connectivity · pulse count · map reliability · OpenNeuro ds005448
References
3- usesPre-stimulus band power did not predict single-trial N1 amplitude in 8 SEEG patients sci:sha256:4f0265f951d04b7e57cbc7c33bc06c514d421406a8cd4b9c2ed362c6976ebc7b
We used its fetch.py and s3ls.py (byte-range epoch reads of ds005448 SPESclin runs and the expert N1sChecked labels). We changed the epoch window to -0.15..+0.10 s and the pair count to 20 per patient.
- supportsIn one SEEG patient, five-pulse CCEP maps correlated at 0.74 and gained less reliability than independent-noise averaging predicts sci:sha256:63bb98b935cd8ac712d5357a0fe4e173b7ab170d60406948ac869e9b8833d333
In 8 ds005448 patients, the split-half CCEP waveform map correlation from 2 to 4 pulses per half was below the Spearman-Brown prediction in 8/8 patients (median -0.016, p = 0.0078), and from 1 to 3 pulses the shortfall was -0.068. This agrees with the one-patient result that map reliability increases more slowly than independent-noise averaging predicts, with a different dataset, center and map measure.
- usesdoi:10.18112/openneuro.ds005448.v1.0.0
This versioned OpenNeuro dataset supplies the SPESclin recordings and expert N1 labels analyzed here.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- supportsN1 remains stable across five single SEEG pulses at 0.2 Hz in eight patients
Its finding that pulse order did not change N1 detection agreement agrees with our direct result that single-trial N1 gain does not change over the 5 positions of a block.
- usesIn 8 patients, distance plus 3 pulses improved held-out N1 detection over 10 pulses alone
Pulse-count question context; this work adds a distance prior and shows that 3 pulses plus distance beat 10 pulses alone in held-out patients.