Three to four SEEG channels on different shafts decoded working-memory load better than adjacent channels on one shaft
We compared equal counts of bipolar SEEG channels sampled from one depth shaft or from different shafts when decoding verbal working-memory load (set size 8 vs 4) in 13–15 patients from one OpenNeuro dataset. For k = 2, 3, and 4 channels, mean cross-validated AUCs were 0.572 vs 0.574 (difference +0.002; Holm p = 0.80), 0.582 vs 0.610 (+0.028; 13/14 patients favored different shafts; Holm p = 0.0018), and 0.595 vs 0.625 (+0.030; 11/13; Holm p = 0.017), respectively. Mean absolute within-class noise correlation was 0.55 for adjacent pairs on one shaft, 0.34 for non-adjacent pairs on one shaft, and 0.25 for pairs on different shafts; non-adjacent same-shaft correlation exceeded between-shaft correlation in all 15 patients (p = 6e-5). All channels yielded mean AUC 0.70–0.74, so 2–4 channels retained only part of the information. These offline results suggest a small benefit from spreading 3–4 channels across shafts, but the study does not separate noise correlation from brain-region diversity. It used one task and dataset, 13–15 patients with 20–42 trials each, and 100 random draws per condition; random channel sets were not optimized.
stereo-EEG · SEEG · working memory · brain-computer interface · channel selection · decoding · noise correlation · AUC · verbal Sternberg task
References
4- usesAll-trial feature selection raised SEEG decoding AUC to 0.84 with random labels sci:sha256:d0f8e2f61f20ff7561d81bf30c1c96cfde2212b5dc53c421ac4d3aefa770c0ad
Archived ds004752 band-power feature files (sha256-checked) are the input data.
- supportsIn SEEG load decoding, noise correlation fell with distance but distance did not predict within-patient pair gain sci:sha256:f794988fce14e59d0bf1cb50181936c4abfb4ff9125feb6585126335c24b93c8
Random k-channel sets on different shafts beat same-shaft sets, and same-shaft noise correlation is higher; same direction as the pair-level shared-contact and noise-correlation result.
- usesIn session 1 of 15 patients, IEDs did not decode memory load; masking raised SEEG AUC from 0.70 to 0.72 sci:sha256:632c406c34e1530eea3bb23516da5054eb6c64dbaa63783d6d4628a34fe21bbb
Original extraction of the bipolar band-power features and trial artifact flags from raw ds004752.
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
The analysis uses archived feature files derived from the verbal working-memory SEEG dataset, with raw-file URLs, sizes, ETags, and SHA-256 hashes recorded in the contribution.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- Three to four SEEG channels on different shafts improved recall decoding over channels on one shaft in 19 patients
[Legacy relation: reproduces] Same channel-draw design, decoder and tests on a different dataset and task (ds004789 word recall); the k = 3-4 gain and the noise-correlation pattern replicate, with a smaller absolute gain.
- supportsFour-shaft placement improved high-gamma motor decoding in eight SEEG patients; three-shaft placement did not
Same SPREAD vs ADJ design; the k = 4 gain also occurs for a focal motor signal in a different dataset (ds004473), while k = 3 was not significant at 70-150 Hz.