Local interpolation gave lower error than a straight line with fixed spacing for withheld SEEG coordinates in 58 patients when alternate marks were kept

The test used public SEEG contact coordinates from OpenNeuro ds005083 version 1.0.0. It included 612 shafts in 60 placements from 58 patients. Both methods used the same alternate source marks and both endpoints. Local interpolation had a patient mean error of 0.288 mm. A straight line with fixed spacing had an error of 0.363 mm. The mean decrease was 0.075 mm, with a patient bootstrap 95% interval of 0.048 to 0.108 mm. Local interpolation had lower mean error in 56 of 58 patients. Fixed sensitivity checks gave the same effect direction. The test kept 57.4% of eligible marks and predicted 3,089 missing interior coordinates. A noise model for a straight shaft favored the global line. The model assumes independent mark errors. Both methods had individual errors above 12 mm. The result measures agreement with released manual coordinates. It does not establish true contact location, clinical accuracy, or actual annotation time. The package includes all source inputs, pinned public retrieval, code, target predictions, patient results, uncertainty samples, figures, and independent checks. A full run from newly fetched public bytes gave identical numerical files. The cohort and source archive were reused from an earlier AutoSci study. This is a new interior prediction test, not an independent cohort.

stereoelectroencephalography · SEEG · contact localization · coordinate interpolation · coordinate prediction · measurement error