A fixed power decoder did not detect previous response side in pre-cue SEEG samples from eight patients

We tested whether SEEG power before fixation could identify the previous left or right response. We used OpenNeuro ds004473, version 1.0.1. The study used 1,953 trial pairs from eight patients with epilepsy. The sampled windows were 2.26 to 7.90 seconds after the previous response. A fixed linear ridge decoder used bipolar power at 13 to 30 Hz and 70 to 110 Hz. Five time blocks and a 30-second train gap gave a mean balanced accuracy of 50.09%. The 95% patient bootstrap interval was 47.33 to 52.70%. A conditional label test gave p = 0.2936, or 0.5872 after correction for two primary tests. Accuracy for the current response was 48.95%. The previous-minus-current difference was 1.14 percentage points. Its interval was -2.96 to 5.69 points. This test did not detect previous side information. It does not prove that the baseline has no response history. The package includes paired features, labels, predictions, null draws, source files, retrieval code, and validation results. This is a secondary analysis of an earlier AutoSci feature package. It is not an independent replication.

stereoelectroencephalography · response history · power decoding · ridge regression · conditional permutation · OpenNeuro