A fixed waveform confidence rule gave 24.9 percent mean site accuracy on accepted later pulses within five SEEG sessions

We tested a fixed confidence rule for stimulation-site labels from public SEEG data. The data were from OpenNeuro ds004457, version 1.0.2. We used 1,923 pulses at 185 sites in five patients. One recording pair was used per patient. At each site, separate pulse sets fitted templates, set a threshold, and tested later pulses. Confidence was the gap between the two best waveform matches at 15 to 100 ms. The threshold was the weighted 75th percentile of calibration confidence. The rule accepted 21.6 percent of test response weight. Mean accepted accuracy was 24.9 percent. A random-selection control with the same accepted site mix gave 22.5 percent accuracy. The mean gain was 2.4 percentage points. Its descriptive 95 percent patient interval was -1.1 to 5.1 points. The exact two-sided patient sign-flip test gave p = 0.25. These results do not establish a reliable gain. A stricter rule in patient 2 gave 71.4 percent accuracy on seven pulses, but the site-matched control gave the same accuracy. The test used later pulses in the same site blocks and sessions. It does not establish use in another session or an effect on behavior. The package contains four figures, all derived inputs, predictions, score matrices, source hashes, code, and retrieval instructions. Raw-file checks and a fresh extraction verified the inputs.

stereoelectroencephalography · SEEG · stimulation-site identification · selective prediction · confidence threshold · waveform template matching · site-matched control · brain-computer interface