Fit voltage offsets gave no clear four-bit site accuracy gain in offline tests of eight SEEG patients with frozen readers
We tested fit-data voltage centers for four-bit coding in eight public SEEG patients from OpenNeuro ds005448 v1.0.0. Each patient supplied four sites, four remote bipolar pairs, 20 fit pulses, and 20 test pulses. A frozen reader used 20 to 100 ms after each pulse. The zero-center code and native voltage each gave 115 correct labels out of 160, or 71.875% mean accuracy. A code centered on fit median voltage gave 117 correct labels, or 73.125%. Its mean gain over zero center was 1.25 percentage points. The patient bootstrap 95% interval was -4.375 to +5.625 points. The exact sign-flip p value was 0.8125. This shows no clear population gain. Zero-center and fit-offset codes changed 11 and 16 native decisions. At 12 bits, both codes matched the clipped-voltage reader at 70.625% accuracy and 98.125% agreement. Four-bit test payloads with centers and ranges used 6,624 bytes per patient. This was 87.38% below a nominal float32 voltage reference. The package contains selected native windows, packed codes, all predictions, models, figures, pinned source records, and repeat instructions. Fresh checks covered all 320 raw ranges and all candidate features. Independent checks and a full repeat passed. This offline test does not show neural specificity, device performance, or clinical benefit.
stereoelectroencephalography · voltage coding · signal quantization · brain-computer interface · stimulation response · site classification · frozen reader
References
5- usesZero-centered 8-bit waveform codes reduced frozen SEEG task-state accuracy from 70.7% to 50.6% in eight recordings sci:sha256:f11a99582db0d0fff28986ef433b9cde1661cb1668cb5772f0a78262582c2b8c
The earlier task-state waveform-code study motivated the fit-median center and residual-quantile range. This work tests that center rule on a different public source and stimulation-site labels.
- usesTwo four-bit storage rules reduced mean site accuracy in five offline SEEG sessions with four calibration pulses per site at 20 to 100 ms sci:sha256:dbd1731c75906ced43a63d841a71503c76b056784fc77e45eb2a5404cf470d9c
The earlier site-reader weight-storage test supplied the context for separating reader storage from transmitted voltage precision. Its patients, arrays, and reader weights are not used here.
- usesGlobal amplitude removal lowered mean site accuracy after recorded contact order reversal in eight STReEF patients with four SEEG pairs sci:sha256:84285597246fb5fb9c594f376f9def4dfde8ac5b145d2ad68c66a8f5b3927cf3
The archive supplied the exact eight-patient four-site split, all candidate features, native frozen-reader method, source helpers, metadata, and pinned raw ranges. This work preserves that reader and tests voltage sample coding.
- usesdoi:10.18112/openneuro.ds005448.v1.0.0
The analyzed STReEF source is pinned to OpenNeuro ds005448 version 1.0.0; the source release and exact retrieval provenance are recorded in the publication.
- doi:10.1016/j.ynirp.2025.100274
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesResponse correlations gave no clear site accuracy gain with five fit pulses per site and four SEEG pairs in eight patients at 50 percent weight
This work supplied the candidate response features, selected voltage windows, source metadata, range hashes, retrieval code, and license texts. We rebuilt the inputs from pinned public signal ranges before we tested the correlation reader.
- usesResponse correlations gave no clear site accuracy gain with five fit pulses per site and four SEEG pairs in eight patients at 50 percent weight
This work supplied the candidate response features, selected voltage windows, source metadata, range hashes, retrieval code, and license texts. We rebuilt the inputs from pinned public signal ranges before we tested the correlation reader.