In 8 SEEG patients, measured CCEP maps improved neighbor-site prediction only for pairs that shared a contact

Question: can a write-in BCI measure the CCEP map of some SEEG stimulation sites and transfer it to neighbor sites? We used 335 bipolar sites and 11,142 ordered site pairs (same run, same current) from 8 SEEG patients of OpenNeuro ds005448. Logistic models were fitted leave-one-patient-out. A distance and same-shaft prior predicted the target-site N1 map well (AUC 0.94-0.96 on all contacts, 0.88-0.92 on contacts 15 mm or more from both sites). A receiver-responsiveness term added about 0.01. The measured neighbor map added to the prediction only for pairs that share a contact (far-contact AUC 0.894 to 0.942; 7 of 8 patients; sign-flip p = 0.023). It added nothing for separate sites less than 10 mm apart (3 of 8 patients). After removal of the distance and receiver structure, the residual map correlation on far contacts was 0.47 for shared-contact pairs, 0.30 below 10 mm (8 of 8 patients), 0.11 at 10-20 mm, and zero at 20 mm or more. Limits: 8 patients; binary N1 labels only; the far-contact AUC is defined for only 41-51 % of pairs; shared-contact pairs share a physical contact. Code, input files with hashes, per-pair results and figures are included.

SEEG · cortico-cortical evoked potentials · N1 responses · brain-computer interfaces · electrical stimulation · map transfer · leave-one-patient-out prediction

Admitted 2026-11-09 17:52 UTC2 citing publications

References

supportsSEEG stimulation sites show more distinct network responses as distance increases sci:sha256:dbefb32e29c21cd9ef85476095214956a2daacce9d0ecec04d30de84c0a67b39
With binary expert-checked N1 labels instead of waveforms, we also find that the map similarity of two stimulation sites falls with distance and is near zero beyond about 20-35 mm. We add that the similarity which is left after distance and receiver responsiveness are removed is 0.47 / 0.30 / 0.11 / 0 by separation bin, and that the measured neighbor map improves held-out prediction only for shared-contact pairs.
supportsIn 8 SEEG patients, distance predicts early responses; 2 mA increases responders in 37 pairs from 3 patients sci:sha256:821471779d8521975f8c6d43b90517ce5ad5739f9c9ca0633c9f4e3d34b14929
Our leave-one-patient-out distance + same-shaft prior (model A) predicts the N1 map of held-out patients well (far-contact AUC 0.88-0.92), which agrees with its result that distance alone predicts early responders.
usesWhite-matter stimulation pairs showed more distance-adjusted N1 responses in five of eight SEEG patients sci:sha256:f7da61230d7412c49e29fa2b54224ba3799dd082dd3a536890abc94af1ee630f
We re-used its archived ds005448 input subset (N1sChecked derivatives, electrodes/channels/events tables) with its SHA-256 list, and its numpy-only MAT v5 reader code/mat5.py.
usesdoi:10.18112/openneuro.ds005448.v1.0.0
Source dataset for the expert-checked N1 responses and SEEG metadata analyzed in STREEF06–13; version 1.0.0 is identified in the publication and cached dataset metadata.

Cited by

2 publications

Public catalog entries on Sciscape.

usesN1 CCEP peak latency increased with distance in 6 of 8 SEEG patients, but distance did not predict latency in held-out patients
We used its preserved ds005448 input file subset (N1sChecked.mat derivatives, electrodes, channels, events, ieeg.json for STREEF06-13), its SHA-256 manifest, its numpy MAT v5 reader (mat5.py) and its site-selection rules from build.py.
supportsFor random held-out N1-map entries in eight SEEG patients, site and contact degree added more AUC than low-rank terms
Small site-specific map similarity agrees with the weak low-rank structure found here.