An early 90 percent mean variance rule left later rest errors above 10 percent at most selected SEEG pairs in seven people

We tested spatial voltage compression in 727 selected bipolar SEEG pairs from seven people. The public source was OpenNeuro ds006065, version 1.0.0. Models used 60 seconds of early rest. A rule that kept 90 percent of total normalized calibration variance left 78.9 percent of pairs above 10 percent squared reconstruction error in later rest, averaged across people. The descriptive 95 percent person bootstrap interval was 67.4 to 90.3 percent. A rule that limited calibration error at every pair reduced the later fraction to 32.3 percent. It retained 68.8 percent of the coefficients, compared with 50.6 percent for the mean rule. The pair rule still left 64.2 percent of pairs above the error limit in the 70 to 100 Hz band. The package contains exact covariance-based scores, source retrieval recipes, hashes, code, models, and four figures. Direct voltage checks and a second analysis run verified the results. Coordinate selection does not prove electrode type. No new artifact rejection was applied. These results concern voltage reconstruction in selected rest recordings. They do not establish a benefit for task decoding, hardware, or stimulation control.

spatial compression · principal component analysis · stereoelectroencephalography · voltage reconstruction · signal processing · brain-computer interfaces