Added SEEG before a response gave no clear key prediction gain over earlier SEEG and task history with a linear reader in eight recordings

We tested whether SEEG before a response improved key prediction beyond earlier SEEG and task history. We used all eight recordings from OpenNeuro ds004473 version 1.0.2. Separate linear readers used early trials to select group penalties and eight pairs per signal window. The groups could be omitted. We tested 791 later trials from 1,970 eligible trials. Added SEEG reduced equal-person mean log loss by 0.00337 nats per trial. The descriptive 95% bootstrap interval was [-0.00739, 0.01654]. The exact two-sided sign-flip value was 0.75. The added input helped two people, harmed one, and was omitted in five. An extra earlier-window control gave a similar small mean gain. This test did not establish an added prediction gain. It does not prove that key information is absent. The package preserves code, inputs, model fits, predictions, figures, and source retrieval records. All 3,164 test probabilities replayed correctly. Independent raw checks passed for 16 trials and 15,012 power values. This is an offline analysis of a reused, quality-selected sample with task changes over time. It does not test continuous response detection or stimulation.

stereo-EEG · SEEG · brain-computer interface · key prediction · log loss · group penalty · chronological validation · epilepsy

Admitted 2026-11-09 17:28 UTC2 citing publications

References

usesLocal baseline subtraction did not establish higher cued-key accuracy with eight SEEG pairs in later trials from eight people sci:sha256:6c99860110fe58ee33309daa23499a12d8a66788bbc4ea390912ae137c28c542
Prior baseline subtraction and concatenation tests informed the new question. We used separate group penalties and an omit-group option to test added prediction value. The study is a related test, not a direct replication.
usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
We reused and verified 121 archived feature, metadata, and retrieval-code files. They supplied the eight SEEG recordings for the new added key prediction test. Fresh raw checks verified two trials per person.
usesFour SEEG pairs gave 56.7% mean balanced accuracy for cued response labels in later trials from eight participants with known response times sci:sha256:9f74e1c598a72b5a49994d79554c68eb3c2f40a881fa70e57a428d2e0883652a
Prior key prediction and earlier-window results motivated the earlier-input control. The known high score for person 2 motivated the planned descriptive exclusion. Our probability-loss test, history inputs, group penalties, and trial split differ.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used the eight recordings from OpenNeuro ds004473 version 1.0.2.
doi:10.1088/1741-2552%2Facae0a

Cited by

2 publications

Public catalog entries on Sciscape.

usesSelection for key identity had lower mean later key error than activity selection in eight offline SEEG records with tuned readers
The prior key study helped define the question and checks for cue timing and trial history. This test uses the same cohort and is not an independent replication.
usesKnown response times gave mean test gains below 0.02% of baseline loss in linear SEEG power models for eight people
Its key prediction result helped select a different question: power prediction with known cue and response times.