Weighted site positions reduced squared location error but reduced 10 mm success in five offline SEEG sessions

We tested a weighted position estimate in public SEEG data from five patients with epilepsy. The test used 137 mapped stimulation sites and 548 separate test pulses. Each site had four fit pulses and four test pulses. The same voltage reader supplied both position estimates. We selected the weight scale from fit pulses only. The mean patient reduction in squared location error was 33.3% against the winning-site position. The mean fraction within 10 mm fell from 55.8% to 36.6%. Both changes occurred in all five patients. Thus lower average error did not imply more close estimates. Recording time alone gave 91.9% success within 10 mm. The blocked schedule limits neural and online claims. This is an offline method comparison in one public cohort. The package preserves voltage inputs, code, source hashes, pulse decisions, fitted models, predictions, and three figures. All 110 raw source files passed hash checks. Raw voltage reconstruction and a fresh-process numerical replay passed.

stereoelectroencephalography · SEEG · brain-computer interface · position estimation · spatial error · electrical stimulation · offline evaluation