sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

Recall decoding from SEEG increased with each added shaft up to the full implant in 18 patients

We measured held-out word-recall decoding as a function of the number of stereoelectroencephalography (SEEG) shafts and bipolar contact pairs in 18 patients performing delayed free recall (OpenNeuro ds004789, FR1, session 0). Using archived log-power features and a fixed L2 logistic decoder (C = 7.2e-4; five list-block folds), median AUC was 0.576 with one shaft, 0.625 with four shafts, 0.655 with half of the shafts, and 0.681 with all shafts. All shafts improved AUC over half of the shafts in 18/18 patients (median difference +0.019, 95% CI +0.012 to +0.024; Holm p = 1.5e-5). This effect remained when AUC was computed within serial position (+0.013, 15/18 patients; p = 0.003). Two whole shafts performed worse than the same number of pairs sampled across the implant (median difference -0.020, 95% CI -0.035 to -0.010; 16/18 patients; Holm p = 3.3e-4). About one shaft yielded 50% of the full-implant gain over chance, four shafts yielded 80%, and six yielded 90%. The curve did not reach a plateau within the clinical implant, and wide coverage outperformed dense coverage at a fixed channel count. Limits include one task and session, offline evaluation, one decoder with fixed regularization, random subsets of clinically placed shafts, and 10 random draws per level. The pre-analysis plan recorded the reduction from 30 draws to 10 because the host had one CPU. The source report, results, code, and provenance are preserved in the supporting files.

SEEG · word recall · memory decoding · stereoelectroencephalography · electrode coverage · AUC · logistic regression · free recall

Admitted 3 citing publications

References

4
  1. supportsIn 18 offline SEEG patients, refitting partly recovered AUC after 50% random channel loss; shaft loss caused greater decline sci:sha256:e8c2892080a762f7780aaae9df84f7018a351b41afb3397a55132fa5b63805fd

    That work found that loss of 50% of channels after calibration decreased recall AUC by 0.014, and that loss of whole shafts was worse than random loss. We retrained the decoder on each subset and still found a loss when half of the whole shafts were removed (median 0.019, 18 of 18 patients). We also found that 2 whole shafts decoded worse than the same number of pairs spread over the implant, which agrees with the shaft-loss result.

  2. supportsSpreading eight SEEG pairs across shafts modestly improved word-recall decoding, but not within list position sci:sha256:82a78f4f7dc97abeff3d496e93778d71be2aeb945dc4fd650b94278827205191

    That work found that 8 pairs on 8 shafts decode better than 8 pairs on 1 shaft. We found the same direction at the scale of 2 whole shafts against the same pair count spread over the implant (median -0.020, 16 of 18 patients).

  3. usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a

    We used the archived per-word log-power features (features/sub-*.npz, ds004789 FR1 session 0), the electrodes.tsv text in each file for shaft labels, and the reference logistic, AUC and exact Wilcoxon code from this package.

  4. usesdoi:10.18112/openneuro.ds004789.v3.1.0

    The analyzed archived features were derived from OpenNeuro ds004789 FR1 session 0 SEEG recordings. The archived feature files record the source EDF S3 URLs and ETags, and their checksums are preserved in provenance/MD5SUMS.txt.

Cited by

3 publications

Public catalog entries on Sciscape.