In 213 SCI publications on 19 public SEEG datasets, reported median decoding AUCs were 0.594 for cognitive targets and 0.765–0.780 for sensory/motor events and CCEP detection

We collected headline area-under-the-curve (AUC) values reported in titles or summaries of SCI publications about public OpenNeuro SEEG datasets. From a snapshot of 484 publications on 2026-10-13, text-mining rules included 213 publications on 19 datasets and excluded 142 by rule. Cognitive-state targets (159 publications, 7 datasets) had median AUC 0.594; 10% reported AUC ≥0.70. Sensory or motor events (26 publications, 6 datasets) had median 0.765, with 54% at or above 0.70. CCEP response detection or prediction (28 publications, 6 datasets) had median 0.780, with 82% at or above 0.70. Publication-level Mann–Whitney tests found higher values for sensory/motor than cognitive targets (p=0.007) and for CCEP than cognitive targets (p=1.1×10⁻¹²). With one median per dataset, CCEP remained higher (p=0.008); the sensory/motor comparison was not significant (p=0.17). Cognitive cross-patient or frozen decoders had a lower median (0.561, n=33) than other cognitive summaries (0.617, n=126). No cognitive dataset had a median above 0.68. This is text mining of author-reported summaries, not reanalysis of recordings. Publications are not independent, contrasts differ in difficulty, and publication bias or selective reporting may affect the distributions.

stereo-EEG · SEEG · decoding · AUC · OpenNeuro · cognition · CCEP · text mining · synthesis