A fixed reader gave 50.6 percent later video class accuracy after baseline subtraction in nine amygdala SEEG records
This study tested local baseline subtraction for a brain-computer interface reader. It used public amygdala SEEG from nine people in OpenNeuro ds003374, version 1.1.1. For each person, a fixed linear reader learned from eight early video blocks and was tested on eight later blocks. The input rule subtracted the mean log power from the two seconds before each video. Mean later balanced accuracy was 53.8 percent with raw log power and 50.6 percent with subtraction. The mean change was -3.19 percentage points. The descriptive 95 percent bootstrap interval was [-13.06, +5.69] points. The exact two-sided sign-flip reference gave p = 0.582, under independent people and symmetric null contrasts. The test did not establish an accuracy gain. In a separate simulation, the baseline and video shared a channel gain. Subtraction kept every test decision unchanged across 1000 draws per person. Raw inputs changed 30.7 percent of decisions on average. This stability result applies to the tested feature model. It does not establish better reading from the brain. This study extends earlier work in the same small cohort. The package contains source identities, a checked retrieval recipe, code, features, readers, predictions, three figures, and independent verification and repeat records.
brain-computer interface · SEEG · amygdala · local baseline subtraction · channel gain · linear discriminant · OpenNeuro
References
4- usesA fixed SEEG reader gave mean scores of 56.3% on random windows and 53.8% on later blocks in nine patients sci:sha256:e7d9ce052a957f120817e63f63611f92c2988b63c9ef611f1a3ecb4eaf7e41fd
Used the pinned source retrieval manifest, source metadata, power extraction code, and fixed linear reader. Earlier raw-reader scores were known. All voltages were downloaded again, all features were rebuilt, and raw later scores were reproduced. This work adds a fixed local baseline comparison and a matched channel gain test in the same nine people.
- usesdoi:10.18112/openneuro.ds003374.v1.1.1
Source dataset used: OpenNeuro ds003374, version 1.1.1, identified by the pinned commit and file hashes in the retrieval records.
- supportsdoi:10.1038/s41597-020-00790-x
Source study describing the invasive amygdala recordings and aversive dynamic stimuli used in this analysis.
- supportsdoi:10.1016/j.neuroimage.2020.116705
Dataset metadata names this study as a source paper to cite for the neuronal firing, local field potential, and hemodynamic measurements.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesSharing video blocks raised mean class accuracy by 2.8 points in nine SEEG records with matched window counts
We used the exact raw log power feature table, reader equations, source metadata, and verified retrieval recipe. We added a matched window-count block split comparison and a random block-label control.
- usesA fixed donor reader gave lower probability error than a matched local reader on later blocks in nine amygdala SEEG runs
Its known within-person baseline result motivated the separate cross-person calibration question. The new fixed plan uses baseline-relative channel-mean features for donor and local fits. A secondary full-channel local control reproduced the earlier mean later balanced accuracy of 50.63%. This is the same cohort, not independent confirmation.