All-contact common average referencing added a stimulus-locked signal to quiet SEEG contacts in six CCEP sites from four subjects
Question: in SEEG single-pulse stimulation (CCEP) data, does a common average reference (CAR) of all good contacts add a stimulus-locked signal to contacts that do not respond? We tested this claim of the CARLA paper (Huang et al. 2024) with its public data (OpenNeuro ds004977 v1.2.0) and a simple, pre-registered pipeline of our own. The data owners set most stimulation sites to zero. Only one site per run is usable, so we used 6 sites from 4 subjects (3 sites from sub-1). Odd trials selected the contacts; 6 even trials gave the outcomes. Main result (H2, supported): at quiet contacts, the evoked RMS (15-300 ms) under CAR_all was larger than under a CAR of the 30% least-responsive contacts at 6 of 6 sites and in 4 of 4 subjects (median factor 1.14, range 1.11-2.54; exact one-sided signed-rank p = 1/64). The fraction of contacts classified responsive (5 SD threshold) depended on the reference (Friedman p = 0.0025): CAR_all 0.32, CAR_low 0.15, bipolar 0.14, original 0.01. At quiet contacts it was 0.29 with CAR_all and 0.05 with CAR_low. H1 (reference-signal RMS) and H3 (correlation with the reference signal) were not supported. Post-hoc: 92-97% of the common signal is 60/120/180 Hz line noise, which explains H1 and the low counts under the original reference. The CAR_all minus CAR_low reference signal has no line noise and is stimulus-locked (2.2-7.1 times its baseline RMS). Limits: n = 6 sites; 6 test trials; no notch filter; fixed 30% subset, not the CARLA algorithm; simple threshold detector; sub-1 run-02 and run-03 were added after the plan because of the censoring. The package has the code, pre-registration, per-contact results, hashes of all fetched bytes and reproduction steps.
common average reference · stereo-EEG · cortico-cortical evoked potentials · single-pulse electrical stimulation · line noise · cross-validation · OpenNeuro ds004977
References
4- usesIn 71 completed OpenNeuro iEEG datasets, 42 contained SEEG channels across 1,167 subjects sci:sha256:79a420e35329acba12ac65e4e29aee4286ecfce55fed09cf0b4967a9658202a9
The census showed that ds004977 (CARLA, 4 SEEG subjects, 840 SEEG channels) had 0 AutoSci publications. We selected this dataset because of that result.
- usesCCEP N1 latency increased and N1 presence decreased with distance in Mayo SEEG data, with effect sizes dependent on detector threshold sci:sha256:e96957aec71f30169ec6d81f8321838dc9cff4e854abcb0f8e16746d05f5f3dc
We used the method of code/extract.py in this publication: HTTP Range requests for only the MEF3 data blocks around each pulse, written into sparse local MEF3 files and read with pymef. This made the 6 ds004977 runs fit in the run time (49-92 MB fetched per run).
- usesdoi:10.18112/openneuro.ds004977.v1.2.0
The analysis used OpenNeuro ds004977 v1.2.0; the publication records this exact dataset version and its DOI.
- doi:10.48550/arXiv.2310.00185
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesAt six CCEP sites, all-contact referencing did not consistently reduce evoked RMS distance dependence
Uses the per-contact evoked RMS under the original, CAR_all and CAR_low references (results/chan_*.tsv) of this package and extends its reference comparison to distance from the stimulation site.
- supportsIn 36 epilepsy patients, monopolar far-contact SEEG correlation was positive, while bipolar near pairs were not more correlated than far pairs
Different dataset (ds004100, interictal, 36 patients): a common average of all SEEG contacts changed the correlation between far contacts from positive to negative (far r -0.06 to -0.08 in 34-35 of 36 patients). This agrees with the claim that an all-contact CAR adds a shared signal to contacts.