A fixed map gave 8.00% mean error reduction for simulated missing centered SEEG voltage at 3–15 ms in 20 calibrated stimulation trains

We tested repair of recorded voltage when SEEG channels were treated as missing. We used the first stimulation train from each of 20 people in OpenNeuro ds005169 version 1.0.0. Sixteen observed channels and all output channels had separate shafts and contacts. Only the observed inputs set pulse edge times. The first 24 pulses fit a fixed spatial map; the next 24 tested it. Each completed 3–15 ms window was centered. The reference was the same two early pulse class means. Mean error reduction was 8.00 (0.86 to 16.63) percent; the interval is a 95 percent person bootstrap interval. The map lowered error in 12 of 20 people. Mean baseline reduction was 9.40 percent. The exact primary sign-flip p value was 0.0646. Matched inputs gave 10.19 percentage points more reduction than wrong-pulse inputs. The result applies to centered recorded voltage with prior output calibration. Shared artifact can aid repair. The test does not show recovery of a unique neural signal or live brain control. The package preserves checked source hashes, native windows, code, fitted maps, estimates, tables, figures, and a clean replay check. These 20 people also occur in the cited prior work; this is a new test in a reused sample.

stereo-EEG · intracranial EEG · missing-channel repair · ridge regression · voltage prediction · cortical stimulation