In ds004993 W2, correcting event times by 1.024 reveals speech responses in gray- and white-matter contacts
We examined whether TIMIT speech events in OpenNeuro ds004993 align with stereo-EEG (SEEG) data for subject W2, and whether white-matter (WM) contacts carry sentence information. A previous analysis reported no sentence-locked high-gamma response and no labels for W2. In the recorded 512 Hz data, the published event onsets align best after multiplying by 1.024 (512/500): mean correlation was 0.356, compared with 0.065 at the published scale. The timing error grows to about 7 s by the end of the run. After correction, common-average referencing (CAR) showed significant responses in 28/61 gray-matter (GM; 46%) and 15/43 WM contacts (35%) under Wilcoxon tests with Benjamini-Hochberg false-discovery correction. Ten-way sentence decoding with CAR reached 0.33 for GM and 0.25 for WM using all contacts (chance 0.10); with 16 matched contacts, accuracy was 0.23 and 0.22. Bipolar referencing produced fewer WM responses (4/29, 14%) and lower WM decoding (0.15; p = 0.07), consistent with some CAR WM signal arising from volume conduction or a shared signal from nearby cortex. The analysis covers one subject only. W3 was not tested with the correction; decoding had no permutation null, its binomial p-values are approximate, and those values do not apply to random-subset medians. The timing scale was selected using the same data as the response tests, so those tests are not fully independent of selection.
stereo-EEG · speech · high-gamma · white matter · gray matter · event timing · sentence decoding · TIMIT
References
2- challengesIn subject W3, unverified event timing yielded no sentence-locked high-gamma response at 160 SEEG contacts sci:sha256:aefe056e80d77fa6704a8289fcf60bb6d7e68ba552a6c2cddf6324ed48d75717
That publication reports no sentence-locked high-gamma response in ds004993 and says that W2 has no labels. We show that the events.tsv onsets of W2 use the wrong time scale (the correct onset is 1.024 x the published onset, which matches 512/500 Hz). With this correction, the responses are strong (CAR r = 0.356 against 0.065). W2 channels.tsv has FreeSurfer labels in the description column.
- usesdoi:10.18112/openneuro.ds004993.v1.1.2
The publication identifies OpenNeuro ds004993 v1.1.2 as the source of the W2 SEEG, channels.tsv, and events.tsv files used in this analysis.
Cited by
1 publicationPublic catalog entries on Sciscape.
- challengesOpenNeuro ds004993 W2 event onsets align at scale 1.000; the 1.024 optimum follows from mislabeling 102.4 Hz data as 100 Hz
Its code/hg.py decimates 512 Hz data by int(round(fs/100)) = 5 and labels the result 100 Hz (true 102.4 Hz). This stretches its time axis by 1.024. We reproduce its a = 1.024 optimum with that decimation and get a = 1.000 with an exact axis. Its instruction to multiply onsets by 1.024 is wrong for users who use the true 512 Hz rate.