An eight-contact common average reference changed fixed task decoder decisions in eight public SEEG records
We tested a smaller common average reference in eight fixed SEEG task decoders. We used OpenNeuro ds004473 version 1.0.1 and kept the eight selected source signals unchanged. The full reference used 119 to 126 good contacts. On 564 later trials, equal-person balanced accuracy was 63.2% with the full reference and 61.1% with an eight-contact reference. The primary loss was 2.1 percentage points, with a descriptive 95% interval from -1.6 to 5.0 points and two-sided sign change p=0.28125. The accuracy loss was uncertain. The eight-contact reference changed 23.7% of decisions on average across people. A matched early fit reached 63.6% accuracy in a secondary test. A shared reference difference trace restored all original decisions in the saved Fourier inputs. This correction requires access to the full reference at acquisition. The package preserves compact inputs, source versions and hashes, all predictions, models, code, figures, and repeat instructions. The study reuses prior trial masks and contact selection. It concerns offline task reading in these records. It does not establish live control or stimulation performance.
stereoelectroencephalography · common average reference · brain-computer interface · task decoder · reference pool · balanced accuracy · SEEG
References
4- usesA median reference reduced mean accuracy loss from simulated white noise at 100 times the training scale on one unused contact in eight SEEG recordings sci:sha256:2df5339eddb5232562539c4891aea2a3649ae2cf867abfbafe410b930c1e9e02
We reused retained trial audits, source good contact lists, early selected contact names, split inputs, source metadata and object versions, EDF retrieval code, and full-reference power inputs for extraction checks. All complete source EDF files were checked again. The new test changes reference membership and tests fixed fits, matched fits, and a shared trace correction. These are the same people and records.
- usesSEEG power before a response gives 57.7% mean balanced accuracy for two cued keys on later trials in eight recordings sci:sha256:9fcbc6aad0f73578b656a03c940b9381ab5791e8f9875ee291d7fc423328ee2d
The inherited trial rules, audits, source metadata, EDF reader, and power method came from this work through the cited reference fault study. We preserve the inherited trial method code. The original task audit enabled fixed eligible trials and source sample windows.
- usesdoi:10.18112/openneuro.ds004473.v1.0.1
The analysis used the public SEEG records and task files from OpenNeuro ds004473 version 1.0.1.
- doi:10.1088/1741-2552%2Facae0a
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesByte parity kept all six-bit SEEG decisions in eight offline records under one known packet loss and correct received bytes
Used its compact Fourier extraction recipe, retained contact order, trial selections, and raw source records through the direct graph parent.
- usesSeven connected difference channels preserved eight-contact SEEG decoder decisions without added error in eight public records
We used its unchanged Fourier inputs, eight-contact selections, trial splits, source records, and extraction code for new graph tests.
- usesLearned eight-contact references reduced reference error, but decoder agreement gains were uncertain in eight public SEEG records
We used its compact Fourier inputs, fixed eight-contact lists, retained trials, time splits, source metadata, and EDF access code. They made a controlled comparison of learned and equal reference weights possible. We checked its complete package and retrieved the complete raw sources again.