One-patient feature reproduction was exact, but controls for the IED-mask decoding gain yielded no decoding result

This package specifies a planned control study of an interictal epileptiform discharge (IED) mask for verbal working-memory load decoding from stereo-electroencephalography (SEEG). The prior analysis reported that removing Welch segments within 100 ms of detected IEDs increased area under the receiver operating characteristic curve (AUC) from 0.70 to 0.72 (+0.018) in 12 of 15 patients, but did not compare with random removal of the same number of segments. The fixed plan compares all segments, IED-masked segments, matched random removal (20 draws), removal of the highest-power segments, and the median across segments. It uses the same imputer, z-score, shrinkage linear discriminant analysis, repeated five-fold cross-validation, trial selection (valid, artifact == 0; set size 8 versus 4), and a two-sided exact patient-level sign-flip test; the primary contrast is AUC(IED) minus mean AUC(RAND). In sub-01, the control extractor reproduced the archived band-power arrays, IED mask counts, and detections exactly (maximum absolute difference 0.0). IED masking removed 1.4% of maintenance segments. Mean absolute log10-power changes from all segments were 0.012 for IED masking, 0.014 for highest-power removal, 0.004 for random removal, and 0.184 for the median. One patient cannot answer the group question. The 15-patient run stopped at its 45-minute work limit on a host with one CPU and load average 50–70: extraction for sub-02 and decoding for sub-01 did not finish. Therefore, this package reports no decoding result and makes no claim about whether the gain is specific to IEDs.

stereo-electroencephalography · interictal epileptiform discharges · verbal working memory · decoding · Welch power · control analysis