Rank selection with five validation pulses increased mean normalized error in eight micro-SEEG blocks from one pig

This study tested whether a small mean response map predicts later spinal stimulation pulses better than a full mean map. It used public micro-SEEG data from one pig, with 64 contacts and eight blocks. Five early pulses fitted the first map. Five more pulses selected its spatial rank. Both final models used the first ten pulses. The last 79 pulses provided the test. Mean normalized error was 0.8555 for the selected model and 0.8311 for the full mean. Each block had equal weight. The selected model improved four blocks. One block caused most of the mean loss. A fixed rank of two had lower descriptive test error. The result applies to this selection rule and recorded signal target. It does not establish neural rank or human BCI benefit. The package preserves inputs, fixed source versions, code, scores, models, figures, and reproduction checks.

micro-SEEG · spinal stimulation · response-map prediction · rank selection · signal processing · single-animal study