Expert raters agree on evoked response labels with median kappa 0.72 in OpenNeuro ds004774
Visual labels of early cortico-cortical evoked potentials (CCEP, N1 response after single-pulse electrical stimulation) are used as ground truth to train and test automated detectors such as ER-Detect. We compared labels in OpenNeuro ds004774 v1.0.0 for 13 rater pairs across 11 patients. The primary analysis used common channel-by-stimulation-pair cells coded 0 or 1 by both raters. Raw agreement was 83–99%, while Cohen’s kappa had median 0.72 (range 0.50–0.88; 2 of 13 pairs below 0.6) and response-label Jaccard agreement had median 0.61 (range 0.39–0.80). The two Mayo SEEG pairs had kappa values 0.70 and 0.83. In a sensitivity analysis counting code 2 as a response, median kappa was 0.67 (range 0.54–0.87). High raw agreement reflects the predominance of no-response cells. Differences in response rates suggest that some disagreement reflects different response thresholds. Detector scores against one rater should be interpreted relative to this inter-rater reference level. The analysis includes only 13 pairs from 11 patients, only two SEEG pairs; codes -1 and 2 are undocumented in the files, and raw signals were not analyzed, so waveform properties underlying disagreement were not tested.
cortico-cortical evoked potentials · CCEP · inter-rater agreement · Cohen kappa · electrical stimulation · OpenNeuro · ER-Detect
References
1- usesdoi:10.18112/openneuro.ds004774.v1.0.0
This study analyzes the derivatives/annots labels from OpenNeuro ds004774 v1.0.0.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesA forward CCEP N1 response predicted higher reverse-response odds in five Mayo SEEG patients, mainly at short distances
This study reuses the MvdB label files, the S3 version IDs and the MAT v5 reader (matv5.py) from this publication. Its inter-rater kappa values set the label-noise limit in our interpretation.