Maps fitted to protect the smaller beta or high gamma fraction raised mean test retention at the same output count in 33 SEEG patients

We tested one fixed spatial map for beta and high gamma SEEG signals. The test used 354 eight-contact blocks from 33 patients with epilepsy. Each patient supplied two separate 60-second records between seizures. A gamma principal component map set the output count from a 90% training target. All maps used that count. We compared a map with equal normalized band weights with a map fitted to protect the smaller retained band fraction. On the separate test file, mean smaller band retention rose from 90.55% to 90.86%. The paired gain was 0.31 percentage points. A patient bootstrap gave a 95% interval of 0.15 to 0.48 points. There were patient gains in 28 of 33 patients and block losses in 113 of 354 blocks. Average beta and gamma retention each fell slightly. The gain persisted with the file roles reversed. At a 95% training target, the interval included zero. The package contains source data recipes, compact covariance data, fitted maps, code, tables, and figures. Four fresh source checks passed. This study reuses a prior cohort. It tests energy preservation and does not establish better task decoding or stimulation.

stereoelectroencephalography · signal compression · spatial projection · principal component analysis · beta band · high gamma · retained energy