Log distance increased mean repeatability prediction error by 1.41 percent at stimulation shafts excluded from training in five SEEG patients

We tested whether a log electrode distance term improves SEEG waveform repeatability predictions at stimulation shafts excluded from training. We used public data from five patients in OpenNeuro ds004457, version 1.0.2. The main sample had 954 links and 147 unordered stimulation pairs. The score was correlation after stimulation minus correlation before stimulation, using two separate five pulse groups. The baseline model had one intercept per recording contact. The second model added log distance. Both models were fitted within each patient. Each test fold excluded an entire stimulation shaft. The equal-patient mean squared error increased from 0.260014 to 0.263676, an increase of 1.41 percent. Distance reduced error in three patients and increased error in two patients. The descriptive patient bootstrap interval for baseline error minus error with distance was -0.024901 to 0.010866. The tested term has no established prediction benefit in this sample. A raw-data rebuild matched all 3,477,760 waveform values exactly. The package preserves the local plan, a coordinate description correction, source metadata and hashes, code, inputs, predictions, figures, and reproduction commands.

stereoelectroencephalography · electrical stimulation · waveform repeatability · prediction error · electrode distance · cross-validation · patient bootstrap