A rule for one known shaft loss gave 39.4% mean worst site accuracy after recorded contact order reversal in eight STReEF patients with four SEEG pairs

We tested whether selection for one known recording shaft loss could improve site identification from public SEEG data. We used OpenNeuro ds005448 version 1.0.0. We kept four stimulation sites in each of eight STReEF patients. Five first pulses per site set the reader and its four recording pairs. Five later pulses tested the reader. The recorded stimulation contact order reversed between these groups. We searched all 2,397,730 four-pair sets. The loss rule gave 39.375% mean minimum accuracy across single-shaft erasures. The intact rule gave 38.125%. The mean difference was 1.25 percentage points. Its descriptive 95% patient bootstrap interval was -6.25 to 10.625 points. Accuracy rose in three patients, fell in four, and did not change in one. With all selected pairs present, mean accuracy fell from 68.125% to 59.375%. This sample did not establish a benefit from the loss rule. The prior pair ranking rule gave 48.125% mean minimum accuracy. The result concerns known feature erasure and response site labels. It does not establish neural decoding or real fault tolerance. The package has code, figures, model states, every test score, and a checked raw-data recipe.

stereoelectroencephalography · SEEG · electrical stimulation · site identification · feature selection · shaft loss · fault tolerance · OpenNeuro · STReEF