White-matter speech high-gamma is largely nonlocal under bipolar rereferencing in two SEEG listeners
We tested whether speech-related high-gamma (HG) in white-matter (WM) stereo-EEG contacts reflects local activity or signals shared with gray matter (GM), and whether the event timing in OpenNeuro ds004993 is valid. We analyzed TIMIT listening recordings from two subjects (W2 and W3), using monopolar, common-average (CAR), and adjacent-contact bipolar references. A lag scan of CAR HG against sentence onsets found peaks at +0.16 s in W3 (top-10 contact r=0.482; z=12.1 versus far lags) and +0.11 s in W2 (r=0.231; z=6.3), with no drift across windows. With CAR, 33/53 W3 and 16/43 W2 WM contacts had significant responses (FDR q<0.05); all-contact speech-versus-silence AUC was 0.959 and 0.940. In W2, sentence-identity accuracy was 0.44 for both CAR GM and WM contacts (chance 0.10; permutation-null 95th percentile 0.15–0.16). Bipolar rereferencing reduced WM-WM speech AUC to 0.694 in W3 and 0.687 in W2, and W2 sentence accuracy to 0.15; GM-GM pairs retained higher speech AUC (0.995 and 0.984) and W2 sentence accuracy (0.29). The results support the interpretation that most WM speech HG in these recordings is shared or volume-conducted, while a small residual WM-WM signal cannot be distinguished from a local signal or a residual GM gradient. The study includes two subjects and one passive-listening run per subject; tissue labels are atlas labels, and no audio-envelope or phoneme model was tested.
stereo-EEG · high-gamma · speech · white matter · gray matter · volume conduction · brain-computer interface · event alignment · TIMIT · OpenNeuro
References
2- challengesIn subject W3, unverified event timing yielded no sentence-locked high-gamma response at 160 SEEG contacts sci:sha256:aefe056e80d77fa6704a8289fcf60bb6d7e68ba552a6c2cddf6324ed48d75717
3f3a001d reported no sentence-locked HG in W3 and blamed the events.tsv alignment. With a lag scan from the neural data, we found that events.tsv is correct: the best lag is +0.16 s, top-10 r = 0.48, z = 12 vs far lags, r = 0.45 at lag 0, and there is no drift over 12 windows. With events.tsv as given, 53/77 GM contacts (CAR) respond at FDR q<0.05. We also found that W2 has per-contact anatomical labels in channels.tsv.
- usesdoi:10.18112/openneuro.ds004993.v1.1.2
The analyzed recordings and BIDS sidecars are from OpenNeuro ds004993 snapshot 1.1.2, as recorded in data_provenance.json.
Cited by
6 publicationsPublic catalog entries on Sciscape.
- supportsOpenNeuro ds004993 W2 event onsets align at scale 1.000; the 1.024 optimum follows from mislabeling 102.4 Hz data as 100 Hz
We independently confirm that the W2 events.tsv onsets align with high-gamma at scale 1.000 and a short positive lag (+0.08 s), with no drift over the run.
- supportsIn one SEEG listener, sentence accuracy was similar with common average and bipolar references without baseline subtraction
Single-channel decoding: CAR white-matter contacts decode sentences, but bipolar pairs outside RPPST13-15 do not; agrees with CAR spread of HG.
- challengesEqual-size white-matter and gray-matter SEEG pair sets had similar key-press AUC in eight subjects
There, bipolar re-referencing removed most WM speech signal; here, bipolar WM pairs alone detected key press above chance in 4 of 8 subjects. Task and dataset differ.
- usesA 20 Hz-locked artifact appears in all contacts of one W3 SEEG run and places lines in the high-gamma band
We use its result that events.tsv timing for W3 timit4 is usable, to label each sentence as VNS on, off or mixed.
- usesIn one SEEG listener, high-gamma best identified sentences with common-average reference, while low-frequency voltage remained informative with bipolar reference
We reused its EDF reader (code/edfio.py) and its finding that ds004993 events.tsv timing is correct. We compare our HG result (CAR 0.49) with its HG-only W2 result (0.44) and extend it with LF and AB features.
- supportsSEEG spectra classify gray- and white-matter contacts in held-out patients, while shaft position does not
That publication found that WM high-gamma under CAR does not stay with a local (bipolar) reference. Our result agrees: WM contacts have much less local power than GM in all bands, and the GM-WM difference is largest with the local shaft Laplacian reference (8 patients, ds005448).