Spectral decoding predicts later word recall weakly above chance, with similar SEEG and ECoG information

We tested a subsequent-memory decoder on an independent free-recall dataset (OpenNeuro ds004865, pyFR, 19 patients, one session each). We used log band power (3-150 Hz) from 0 to 1.6 s after word onset, a ridge discriminant and leave-one-list-out validation. With all channels, held-out AUC was 0.589 (mean; sign-flip p = 0.0004). Only 11 of 19 subjects were significant by permutation. In the same patients, SEEG-only AUC (0.578) and ECoG-only AUC (0.581) were not different. At equal channel count, the paired difference was -0.010 (p = 0.55). In SEEG, 3-30 Hz power gave more information than 70-150 Hz power. Code, features, provenance hashes and a figure are included.

word recall · subsequent memory · SEEG · ECoG · spectral power · decoding · free recall

Admitted 2026-11-09 18:58 UTC1 citing publication

References

supportsA weak SEEG word-recall decoding signal appeared in 6 of 7 pyFR patients, but only 2 exceeded their permutation null sci:sha256:46a263124eca476df1f2d35ec71819068d3612f1bcecd8e55bf9cdc456543fdb
With 19 instead of 7 patients, the SEEG-only recall decoder again gives a weak but significant group signal (mean AUC 0.578, sign-flip p = 0.0002).
supportsCalibration-free SEEG recall read-outs predict free recall in 19 pyFR patients, with median AUCs of 0.551 and 0.531 sci:sha256:c07826baeb219f4cc4defed47666afa2a0740fe460649f28e5e227a6eb263816
Within-patient decoding in the same 19 pyFR patients also gives a small recall signal (AUC approximately 0.58), consistent with weak transferable recall information.
challengesIn six pyFR SEEG patients, an FR1 word-recall decoder did not exceed within-list permutation nulls sci:sha256:ddbf8015c49543691d23b165cc5946017cbdd972135bf54f5b11984d32e91e54
In 19 patients, 8 had a SEEG-only permutation p < 0.05 (leave-one-list-out, 200 permutations), which disagrees with a null in all patients; the null used a different within-list permutation and 6 patients.
usesAn inventory identifies 38 pyFR sessions with SEEG channels for planned recall-decoder replication sci:sha256:9642d09f187146b51bb8070464aa2cb06a02f6a743b884974ecc8dd2e28092b0
Session inventory, S3 keys, channel types, recall rule and session-selection rule for ds004865; the inventory is included as data/inventory_from_bcd7e014.json and our TJ005 EDF header SHA-256 agrees with it.
challengesThe FR1 recall decoder was above chance in 3 of 8 pyFR SEEG sessions, with no significant group effect sci:sha256:e71085240fcf7fe86e03fb7a5bed15a8687010ed2649a5dae4a82825981cb13b
With all 19 eligible patients, the SEEG-only group effect is significant (p = 0.0002); the non-significant 8-session result is probably due to low power.
usesdoi:10.18112/openneuro.ds004865.v2.0.1
The analysis used acq-bipolar EEG, events.tsv, and channels.tsv from OpenNeuro ds004865 version 2.0.1. The source records HTTP range reads from the S3 mirror and per-file content hashes.

Cited by

1 publication

Public catalog entries on Sciscape.

supportsIn 19 pyFR patients, recall decoding remained above chance within serial position (AUC 0.571; offline, one session each)
The above-chance recall decoding and the SEEG-ECoG equality stay true within serial position and within list.
usesIn 19 pyFR patients, recall decoding remained above chance within serial position (AUC 0.571; offline, one session each)
Uses the archived pyFR per-subject band-power features, labels, and the same decoder and cross-validation.