Four-pair monitoring lowered error versus a five-pulse template but not zero for centered 10 ms SEEG voltage at 20–500 ms in eight patients

We tested a four-pair voltage monitor in public STReEF SEEG data from eight epilepsy patients. The test used 96 conditions and 6,803 bipolar pair instances. Five early pulses fit the model. Five later pulses with reversed recorded contact order tested it. Each target shaft was hidden during test inference. Four pairs on four other shafts supplied current voltage inputs. Calibration pulses selected the pairs and ridge penalty. We scored centered 10 ms voltage means at 20 to 500 ms. Mean patient error fell 4.26% from the five-pulse template. The descriptive 95% patient bootstrap interval was 2.77% to 5.73%. All eight patients had lower error. Correct pulse pairing gave lower error than wrong pairing. A similar gain occurred before each pulse. The after-minus-before matched gain was 0.0021, with an interval of -0.0259 to 0.0337. The monitor mean error was 1.3735; zero centered voltage gave 1.3560. Thus the template gain did not establish improvement over zero or a response-specific benefit. This is completed-window reconstruction, not a forecast or a live BCI test. The package preserves complete analysis inputs, native audit bytes, models, scores, figures, source versions, code, and repeat instructions.

stereoelectroencephalography · voltage monitoring · ridge regression · brain-computer interface · epilepsy