Near SEEG Stimulation Sites Were Less Separable in Two Patients, but Distance Predicted Accuracy in Only One
A brain-computer interface (BCI) that writes information with stereoelectroencephalography (SEEG) stimulation needs sites whose responses can be distinguished. We measured single-trial response separability against distance between stimulation sites in two patients from OpenNeuro ds004457 v1.0.2. The data were 6 mA biphasic single-pulse recordings, with about 10 trials per site. A leave-one-trial-out nearest-centroid decoder classified each site pair from 15–295 ms responses. The preregistered one-sided distance–accuracy rank test was positive in sub-1 (rho = 0.19, Mantel p = 0.026) and negative in sub-3 (rho = -0.08, p = 0.82). Descriptively, pairs less than 5 mm apart were separable in 36% and 65% of cases, respectively, compared with 90% and 97% for pairs 20–40 mm apart. Separability increased up to 10–20 mm, did not increase beyond 20 mm, and declined slightly at 40 mm or more. The distance-bin pattern was not preregistered as a statistical test. The results do not establish a general monotonic relation between distance and accuracy. Interpretation is limited to two patients, one run per patient, unequal recording-contact counts (20 and 6), a simple decoder, and possible stimulation artifact in the response window. The analysis was offline, with no closed loop or behavioral outcome.
stereoelectroencephalography · electrical brain stimulation · brain-computer interface · single-trial decoding · response separability · electrode distance · intracranial electrophysiology
References
3- supportsIn one patient, 12 SEEG contacts identified cortical stimulation sites at 25–31% accuracy after 100 ms sci:sha256:503b7b0bad0fdae4a545d917acb95082993a163d006752a9a5d862dda1df1ec5
43c3937b found in one ds003708 patient that near sites give similar single-pulse responses (Mantel p = 0.001). We test the same idea in a new dataset (ds004457) with a pairwise decoder. The support is partial: the pre-registered distance-accuracy rank test is positive in sub-1 (p = 0.026) and negative in sub-3 (p = 0.82), but in both patients pairs less than 10 mm apart are separable less often than pairs 20-40 mm apart.
- supportsWithin-patient SEEG sites under 10 mm apart showed excess percept-quality agreement, but effects varied across patients sci:sha256:e76ad5e2ca09e76ffd855e774cf7dc4efeb66b261150d517091e9c7d4a8ffd42
9a049030 found similar percepts for stimulation sites less than 10 mm apart (ds005169). In ds004457 we find, with a different measure (single-trial response separability, not percepts), that site pairs less than 10 mm apart are separable less often (36-82%) than pairs 20-40 mm apart (90-97%) in 2 patients. The support is indirect because the measures differ.
- usesdoi:10.18112/openneuro.ds004457.v1.0.2
This version of OpenNeuro ds004457 is the recorded source dataset analyzed in this publication.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesOne repeated SEEG site retained identification accuracy across 50 minutes, while cross-site timing effects remained anatomically confounded
We use the single-pulse response features and the provenance files of this work.
- usesIn two SEEG patients, greedy selection reached 90% of all-contact single-pulse site information with 3-4 contacts
The analysis uses the binned single-trial features and provenance files of this package without change.
- usesIn two SEEG patients, 10–13 response-selected sites carried about 2 bits per pulse, more than all sites in 50/50 offline test splits
We use the archived binned single-trial response features of ds004457 sub-1 and sub-3 from this package as the only input.
- usesIn two SEEG patients, a 15–95 ms window retained 88% and 105% of 15–295 ms site accuracy, but the 90% claim failed in one patient
Uses the archived single-trial features (ds004457 sub-1 and sub-3) and the feature code of this publication.