Later input gave a small, uncertain error reduction for five-pulse SEEG forecasts at 86 sites in eight people
We tested whether an approximately 80 ms SEEG input ending at 250 ms after stimulation improved forecasts of the same 300–500 ms response compared with an input ending at 100 ms. At 86 selected sites in eight people, the five-pulse ridge model gave a 1.21% equal-person mean error reduction (95% person bootstrap interval −0.42% to 2.59%). The matched before-pulse control gave 0.65%; the after-minus-before difference was 0.57 percentage points (95% interval −1.19 to 2.36). All intervals include zero. These results do not establish a mean timing benefit or an added stimulation-specific effect for this fixed model. The later forecast reduced error by 0.77% relative to the fixed reference (95% interval −0.18% to 1.79%). Results are exploratory and apply to one session per person, retrospective good-trial labels, and a strong-response readout selection. No live interface, task decoding, stimulation control, or safe setting was tested. The publication includes the scientific report, methods, figures, inputs, tables, forecasts, and verification receipts; the GPL analysis and extraction code is omitted under SCI's MIT code terms, limiting fresh reproduction from this package.
SEEG · electrical stimulation · forecasting · brain-computer interface · ridge regression · timing · epilepsy
References
4- usesDonor response maps increased mean late SEEG forecast error by 0.24 percent in eight patient tests with five local calibration pulses sci:sha256:3a42561d5beee0eb95219f0bd2046baf55311d231a86aeaa677257c56a17ca50
We adapted the local ridge update and the scaled five-pulse target mean. We tested new input times with a common late target.
- usesGain fits with shifts up to 19.5 ms left 1.93 times the matched control variance in 59 to 460 ms SEEG at 86 selected sites in eight people sci:sha256:e12c7736723be38135302ccdbf9a63fa33886bbb1b8b5746fa0c9430d88dfb38
We used the source-checked pulse windows, corrected event order, source pins, and fixed five-pulse readout choices.
- usesdoi:10.18112/openneuro.ds004696.v1.0.1
Source SEEG dataset, version 1.0.1; recorded dataset metadata identifies the version and CC0 terms.
- doi:10.1523/JNEUROSCI.2201-22.2023
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesA calibration waveform reduced mean relative six-bit code error at 86 selected processed SEEG sites in eight people with known pulse times
We used its 16 source-checked voltage and pulse files, corrected source event mapping, fixed five-pulse readout choices, source metadata, and public retrieval recipe. All copied data hashes matched its archived manifest.
- usesEarly signals from other shafts raised mean 120 to 500 ms SEEG forecast error at ridge penalty 1 in eight patients
This publication supplied context for testing early-to-late SEEG forecasts with five local calibration pulses. Its selected strong responses, fit design, and target interval differ. The present test uses metadata-selected readouts and other-condition transfer; it does not reproduce that result.