Local differences and group contrasts preserved common-average voltage in 19 SEEG records

We tested a voltage preservation method on 19 public SEEG records from OpenNeuro ds003876 version 1.0.2. The method kept 1,742 local contact differences and added 265 orthogonal contrasts of group means. Each record needed 9 to 22 added channels. All common-average voltage modes were preserved. The test covered 2,026 contacts and 120 seconds per record. The largest waveform relative error was 1.45e-15. A proof shows that c-1 added linear channels are necessary for c separate contact groups. For fixed local rows and a total added squared row norm of c-1, the contrasts give minimum reconstruction error. This error result requires independent output errors with equal variance after the transform. The mean expected error was 74.66 percent lower than a specified first-contact chain. It was 48.82 percent lower than a specified middle-contact star. Contact errors before the transform remained identical after decoding. Counterexamples show the limits for unequal and correlated output errors. The package includes the tool, proof, coefficients, matrices, results, source pins and repeat instructions. A separate repeat from downloaded source ranges reproduced all numerical tables. No clinical effect was tested.

stereoelectroencephalography · SEEG · common-average reference · signal preservation · bipolar montage · linear transforms · reconstruction error