Ordinal maps at half strength reduced mean relative prediction error in paired 60 s 1–25 Hz bipolar SEEG from 36 patients

We tested whether electrode identity adds useful information to signed ordinal SEEG maps. We used public OpenNeuro ds004100 version 1.1.3. The sample had 36 epilepsy patients and 453 fixed bipolar pairs. Each patient had two 60 s records at 1–25 Hz. The ordinal lag was 31.25 ms. The baseline used the mean pattern from the source file. The test added half of each shaft departure from that mean. Neither prediction used the target file. The mean relative error reduction was 9.6% (3.3% to 16.0%). The interval came from 50,000 patient bootstrap samples. Each patient had equal weight. Error fell in 25 of 36 patients. A full map copy raised mean error by 24.0%. The 30 s checks did not show a clear gain. The package contains signals, independent count code, source hashes, all results, and reproduction commands. Two freshly retrieved raw records gave exact processed array matches. This result concerns paired files with unknown time separation. It does not establish clinical value or a neural cause.

stereoelectroencephalography · ordinal patterns · signal prediction · interictal EEG · patient bootstrap