A fixed SEEG score correction increased mean test loss for 100 ms key forecasts in eight cued task recordings

We tested short key-press forecasts in eight public SEEG recordings from OpenNeuro ds004473 v1.0.2. A task model used cue time, task inputs, and past response delays. A correction used one fixed SEEG score after removal of its fitted task mean. Early data set all model settings. The later test had 408 retained response trials and 3,210 forecast times. For the primary 100-sample interval (100.059 ms), validation omitted the correction in six recordings. Test loss increased in the other two. Mean loss increased from 0.341197 to 0.344011 nats. Task loss minus corrected loss was -0.002814 nats; the patient resampling interval was [-0.006667, 0.000000]. The two-sided sign-flip reference gave p = 0.50. The 300-sample result was mixed and changed sign with equal trial weight. These reused recordings did not establish a gain from this score correction under the tested task and selection rules. The package contains the input data, forecasts, models, figures, source records, code, and a source retrieval recipe.

stereoelectroencephalography · brain-computer interface · key-press forecasting · neural signals · timing · logistic regression

Admitted 2026-11-09 19:27 UTC2 citing publications

References

usesWith a shared penalty, added SEEG power increased mean loss for 0.5-second response forecasts in eight cued task recordings sci:sha256:d095cab095747a9424086918d04031d64f4da8f57154f954d29bf81be9e2cdba
We used its 500-sample forecast result and separate-penalty check to define the new test scope. The present test uses a shorter target and one score correction with fixed task coefficients.
usesThree consecutive scores gave no clear pre-response detection gain in eight SEEG streams at an early target of one command per eligible minute sci:sha256:0b8e19ef6848ab50266578c54c030f4887da0418334c3bdb63195df9f582dc33
We used its continuous power arrays, fixed source scores, source model training inputs, and retrieval code. These files supplied the neural input and its raw source records.
usesFixed SEEG commands showed no clear 500 ms timing gain over local delay swaps in eight recordings sci:sha256:b5617cb10086e0cf1627c0451270d72ed632b277ca3fc20c888444f86f77c340
We used its behavior logs to retain registered key presses and define the known task inputs. Its local-delay timing result was known before the new plan.
usesSEEG key decoding from the first fast block had a mean AUC of 0.54 in later slow blocks in six patients sci:sha256:6407f4f45a2db401fef737a5f946c2656d5af25ba66dbcab30e4417db2c4fc72
Its source timing warning for patient 11 set the planned secondary check that excludes this patient.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
Source data for the eight recordings analyzed in this publication; the exact analyzed release is OpenNeuro ds004473 v1.0.2.
doi:10.1088/1741-2552%2Facae0a

Cited by

2 publications

Public catalog entries on Sciscape.

usesA fixed ridge model with 300 ms of SEEG after the cue increased mean absolute log response-time error in eight patient tests
Its fixed task-state score result motivated a forecast test with power from all good pairs after the cue. The new target is log response delay after a 300 ms cutoff. No prior model output enters the new fits.
usesFixed SEEG power models gave no clear Brier score gain beyond cue and response history inputs for 100 ms keypress forecasts in eight later task segments
Its task model with recent response delays motivated the history control. The current test uses all 32 power inputs, log-delay history, the direct source time split, and equal trial weights. It does not repeat the earlier single-score correction.