In 18 SEEG patients, within-list recall decoding had median AUC 0.659 vs 0.669 overall; list-centering did not improve it
Closed-loop memory stimulation decides word by word, so a recall decoder must rank words within a list, not only follow a slow list-level state. We evaluated the W_chan decoder on archived per-word band-power features from OpenNeuro ds004789 (FR1 free recall, session 0) in 18 patients. The plan preceded the analysis. The within-list AUC had a median of 0.659 (17/18 above 0.5; exact one-sided Wilcoxon p = 7.6e-6), compared with a median overall AUC of 0.669; the paired difference was +0.014 (11/18, p = 0.15), which does not establish equivalence. The median retained fraction was 1.06. List-mean output had median AUC 0.519 (12/18, p = 0.019), despite 14% of output variance being between lists. In 16/18 patients, list recall rates did not vary more than in a label shuffle. Removing each list's feature mean did not change within-list AUC (median difference -0.006, Holm p = 0.15). Causal running-mean normalization decreased it (median difference -0.019, 4/16, Holm p = 0.022). One patient had a strong list-level component: within-list AUC 0.575 vs 0.689 overall. The results support checking within-list AUC per patient before using word-level triggers. This offline analysis covers one session per patient; a non-significant difference is not proof of equality. The package includes the plan, code, archived feature inputs and hashes, results, figure, and reproduction steps.
SEEG · free recall · memory decoding · within-list AUC · closed-loop stimulation · intracranial EEG · cross-validation · feature normalization
References
4- supportsIn 18 ds004789 FR1 SEEG patients, previous-word decoder output weakly predicted next-word recall, while averaging consecutive outputs reduced recall ranking sci:sha256:c0dbde9e1a4edbfea60de10c703c77df04d4378d11673a3b83e57b852130cf12
Our decomposition shows that the list-level part of the same-word decoder output carries little recall information (list-mean AUC 0.519). This agrees with its finding that the small next-word prediction is list level.
- supportsIn 18 SEEG patients, word-n recall-decoder output weakly predicted next-word recall after serial-position control sci:sha256:43f07e9a2c99c278c2a8ceb8873b75e6cbd5a6a410f0b3d73524bc8f2e940d0a
Our result agrees with its finding that the shared state of adjacent-word outputs is mostly slow and list level, and that the same-word information is item level.
- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
We used its archived per-word band-power features (copied with MD5 sums), its W_chan decoder settings, its inclusion rule and its helper code (logistic regression, AUC, exact Wilcoxon).
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
The archived per-word features are attributed to OpenNeuro ds004789, FR1 session 0; the saved version-pinned dataset record identifies v3.1.0. The feature arrays and their MD5 hashes are included in the package.
Cited by
1 publicationPublic catalog entries on Sciscape.
- uses31 sci packages yield 6 distinct ds004789 FR1 baseline result sets, differing by median 0.044–0.066 AUC per patient across 17 patients
We used column AUC_total of results/per_patient.tsv as the baseline AUC (family F1).