A fixed ridge penalty increased beta squared coherence transfer in 9 of 19 selected SEEG records

We tested a fixed ridge penalty for transfer of complex canonical beta coherence weights to the next minute. The source was OpenNeuro ds003876 version 1.0.2. The selected sample had one interictal SEEG record from each of 19 patients. We fit weights on 30 Fourier segments and tested them on the next 30 segments. Each segment was 2 seconds long. The penalty was alpha 0.1 times mean training channel energy. Each patient had equal weight. Ridge increased the mean test squared coherence in 9 patients and decreased it in 10. The mean gain was 0.004035. The 95 percent patient bootstrap interval was -0.003578 to 0.015681. The exact two-sided sign reversal p value was 0.710533. These uncertainty measures require independent patients and the stated sampling and null assumptions. The test did not show a consistent patient increase or establish a zero effect. Fixed penalty checks and 32 phase control draws per patient are preserved. All 19 fresh source ranges reproduced the archived Fourier coefficients exactly. A separate solver and an independent full repeat passed. The package preserves inputs, code, source hashes, cell scores, weights, patient results, figures, and reproduction instructions. The scope is adjacent minutes in this selected beta-band sample. The result does not establish anatomical connections or clinical accuracy.

stereoelectroencephalography · SEEG · canonical coherence · ridge regularization · beta band · signal connectivity · transfer testing · OpenNeuro