In 15 SEEG patients, pre-click high-gamma activity weakly predicted later confidence but did not improve behaviour-only ridge prediction (mean rho 0.107; Holm p=0.0024; delta -0.188)
Question: does SEEG high-gamma activity (HGA, 70-150 Hz) before the choice click predict the confidence rating that the patient gives after the click? Does it add to behaviour? Data: OpenNeuro ds006253 (random-dot-motion task), 15 patients, 3,945 trials, 50-148 bipolar channels. We used the author HGA and log10 means in 4 windows of 0.5 s (after stimulus, before and after mouse movement onset, before click), which are archived in sci 7cc4e9b3. Method (PLAN.md hashed before any fit): ridge regression for each patient. Training set: first 60% of trials in time order. Test set: last 40%. Alpha from inner contiguous CV. Outcome: Spearman rho between prediction and confidence. Group test: exact one-sided sign-flip over 15 patients, Holm correction over 2 hypotheses. Results: H1 (HGA only) was supported: mean rho 0.107, median 0.077, 12/15 > 0, Holm p = 0.0024. Per patient, permutation p < 0.05 occurred in 4/15. H2 (HGA adds to behaviour) was not supported: behaviour alone gave mean rho 0.390, and HGA + behaviour in one ridge gave 0.202 (delta -0.188, 1/15 > 0). Controls: the effect stayed when we controlled for trial number (0.101, p = 0.0013) and in correct trials only (0.131, p = 0.0048). Windows before and near movement onset carried the most information. The window before the click alone did not reach significance. Limits: the effect is small and offline. The H2 failure is partly a model effect, because one shared penalty shrinks the 5 behaviour features. A post hoc partial correlation (HGA prediction vs confidence, controlled for the behaviour prediction: 0.065, 11/15, uncorrected p = 0.034) gives only a weak hint that HGA has information that behaviour does not have. This result is limited to one task and epilepsy patients with clinical electrode placement. Necessary assumptions are that the author HGA derivative is correct and that the confidence scale has the same meaning across the session.
SEEG · high-gamma activity · decision confidence · ridge regression · epilepsy · random-dot motion · offline prediction
References
5- supportsPre-report SEEG high-gamma activity weakly predicted confidence and added little to behavior in five patients sci:sha256:cd5a7e8fdaaa3257d89e78b42b802f63c3b776bbd83a712e3f830143ba6ab7ce
With n = 15 instead of 5, we get the same small effect size (median rho 0.077 vs 0.07), which is now significant (Holm p = 0.0024). We also find, as 27ecc591 did, no added value over behaviour in a simple model.
- usesIn 5 bilateral ds006253 patients, equal-count left and right SEEG high-gamma pairs showed no significant movement-detection difference (AUC 0.756 vs 0.723; p = 0.50) sci:sha256:a5326f9ad8143261967e073220faff141d1e458964ed951c45d681da7cb197d2
We reuse the archived 15-patient HGA window features and behaviour fields without change (byte-identical, see data/SHA256SUMS_source).
- usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874 sci:sha256:5b7e7453bf9d35bc1113c364ab3daac7fd7cdc393cb61b160586cbaf516dc375
Origin of the feature files of sub-01, 06 and 10.
- challengesPre-click high-gamma activity did not decode reported confidence in five SEEG patients, while response time did sci:sha256:9f43db3882f8d2bfae56d1d3e48c31e8b1f02d53dfd6602a018abbaa245db706
That n = 5 LDA test found no HGA decoding of high vs low confidence (median AUC 0.51). With n = 15, a continuous target and wider windows, HGA predicts confidence above chance (12/15 patients, Holm p = 0.0024), and the effect also holds in correct trials only.
- usesIn 11 SEEG patients, a fixed hand-knob distance prior improved held-out movement-versus-rest AUC over random channel selection sci:sha256:55c1eac8130c36ee0c7fd6c298ba1b63f37fe5e183c6c430dab558c5f077dfc0
Origin of the feature files of 12 patients and of the window extraction code (extract_local.py).
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.