At factor 3.4, the 50 µV baseline minimum increased precision and decreased sensitivity for DvB N1 labels in two ds004774 SEEG records

The 50 µV baseline standard deviation minimum changed SEEG N1 label agreement in two records from OpenNeuro ds004774 version 1.0.0. We fixed the threshold factor at 3.4 and kept the source peak rules. The primary test used 438 cells from 12 SEEG contacts during surface stimulation. The reference was binary manual labels from reviewer DvB. Removal of the minimum added 55 calls. Of these calls, 13 had an N1 label and 42 had a no-N1 label. In UMCU21, sensitivity rose from 11.3% to 16.9% and precision fell from 61.5% to 57.1%. In UMCU25, sensitivity rose from 32.6% to 53.5% and precision fell from 93.3% to 37.1%. Balanced accuracy rose by only 1.24 and 0.82 percentage points. With binary SB labels in UMCU21, precision rose from 20.0% to 21.1% after minimum removal. These two patients do not establish a population effect. Cells share contacts and stimulation pairs. The archive preserves exact SEEG signal windows, source labels, version hashes, code, results, figures, and reproduction instructions. The detector exactly reproduced stored calls, amplitudes, and peak indices for all 480 retained cells. A separate scalar check verified all 12 result rows. A full repeat produced identical data, tables, and figures.

SEEG · electrical stimulation · evoked responses · ER-Detect · threshold · N1 · sensitivity · precision · manual labels · OpenNeuro