Equal epoch weights did not show a clear lower-tail high gamma gain at matched output counts and a 90 percent training target in 33 SEEG patients

This study tested whether equal weights for one-second epochs improve high gamma power retention in a separate SEEG file. It used OpenNeuro ds004100 version 1.1.3, 33 patients, 354 blocks of eight good contacts, and 60 seconds from each of two interictal files. Both maps used the same output count. The count came from a 90 percent total-power target in the training file. The primary measure was the paired change in the test epoch 10th percentile, with blocks averaged within patients and patients weighted equally. The mean change was +0.025 percentage points. Its 95 percent patient bootstrap interval was -0.013 to +0.061 points. The two-sided patient sign-flip p value was 0.2294. This result did not establish a positive mean gain or equivalence. Total test retention was 91.949 percent for both maps. Of 294 blocks that passed a 90 percent total-power test, 124 had an epoch 10th percentile below 90 percent. Some fixed secondary checks had small positive intervals. They do not replace the primary result. The package adds epoch covariance data, map weights, four figures, complete code, a frozen local plan, and pinned source-byte recipes. All 66 source prefixes passed hash checks. Rebuilt means reproduced the prior covariance and total-power baseline. Separate projector, raw epoch, and fresh-process checks support the calculation. No task decoding, live BCI, device power saving, or brain stimulation benefit was tested.

stereoelectroencephalography · SEEG · high gamma · spatial compression · epoch retention · principal component analysis