Diagonal covariance models retained 51% of top shaft links after common average reference in paired 2–40 Hz SEEG samples from 36 patients

This study used paired public SEEG samples from 36 patients in OpenNeuro ds004100, version 1.1.3. The question was whether common average reference can retain top shaft link ranks without covariance between distinct input contacts. An exact model kept measured contact variances and set all other input covariance entries to zero. For the 2 to 40 Hz pass band, the model retained 51.0% of its top 10% shaft links across recordings. Controls that shuffled second-record variance labels retained 10.8%. The mean patient difference was 40.2 percentage points, with a 95% patient bootstrap interval of 30.1 to 50.2. The shuffle contrast measures persistent variance assignment across the shaft layout. The selected model correlation scores were small. The model shared only 14.7% of selected links with recorded-signal graphs within a record. Thus, stable ranks can occur in this model without input covariance between contacts. This result does not identify false links in acquired neural signals or show independent neural sources. The package preserves code, inputs, controls, results, figures, source versions, and instructions. Fresh pinned downloads rebuilt all 72 inputs exactly. A second analysis run gave identical result files.

stereoelectroencephalography · SEEG · common average reference · covariance · rank stability · signal processing · epilepsy