A distance model gave 1.1% higher mean relative 20–100 ms SEEG response error on whole held-out shafts in four patients

We tested whether anatomical distance can reduce stimulation calibration at a new electrode shaft. We used public SEEG data from OpenNeuro ds004457 version 1.0.2. The fixed metadata rules left 115 sites on 28 shafts in four patients with epilepsy. A fifth patient had only two eligible shafts and was excluded by the fixed three-shaft rule. Ten fixed bipolar pairs measured the responses. Each outer fold removed all signals from one stimulation shaft during fitting and parameter selection. Inner shaft folds selected a Gaussian distance width and a weight on the global mean. Calibration used the first four good 6 mA pulses at each donor site. Testing used the final four pulses at each target site. The primary endpoint was mean squared centered voltage error from 20 to 100 ms. The distance model gave 1.05% higher mean relative error than the global mean response. The descriptive patient bootstrap interval for error reduction was -4.60% to 1.89%. Four local target pulses gave 81.47% lower error. A time model gave 1.59% higher error. A fixed check with recording pairs that share no contact gave 3.13% lower distance error. Thus the result depends on the recording set. It does not show that distance cannot help in general. The package preserves input bins, source metadata, a dated plan, complete folds, predictions, figures, code, and a pinned raw retrieval recipe. Independent error checks, held-shaft signal replacement, a full repeat run, and raw reconstruction were used for verification. The study is an offline test in four sessions. It does not establish a prospective controller or a perceptual effect.

stereoelectroencephalography · electrical stimulation · response prediction · anatomical distance · calibration · whole-shaft validation