SEEG response shapes at 15–100 ms gave higher site identification accuracy than shapes at 100–500 ms within each of five patient sessions

1. usesFour early pulses selected larger later responses at five SEEG contacts within each recording with the source reference sci:sha256:85335f12c83ace7447d3fe10824dee06a620ddfcf0f8511fddb5b8815e505409

   We use this report as related work to distinguish site selection by voltage magnitude from identification by single-trial waveform shape. We inspected it after our decoder results were complete. Both studies use the same source patients; this is not an independent patient replication. We imported no data or code from that report.

2. usesdoi:10.18112/openneuro.ds004457.v1.0.2

   The analyzed SEEG recordings and stimulation events are from OpenNeuro ds004457 version 1.0.2.

3. doi:10.1523/JNEUROSCI.1325-22.2023

Public catalog entries on Sciscape.

• usesFour learned coefficients gave higher site accuracy than four fixed cosine coefficients for later pulses in five SEEG sessions at 20 to 100 ms

  We used its fixed recording pairs, event rules, chronological splits, feature arrays, metadata, and raw retrieval and extraction code. We rebuilt all feature arrays from the public raw files before the new compression comparison. This is a dependent method study in the same five sessions.

• usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions

  We used this earlier site benchmark to choose the pulse-site identification question and response-window controls. It uses the same source patients. We did not treat its evidence as an independent sample.

• usesFour sites selected from SEEG shapes at 30–100 ms gave higher mean pulse identity accuracy than random sets within five patient sessions

  We used the trial signal bins, source metadata, fixed recording contact pairs, and extraction and retrieval code. We verified their archive hashes and rebuilt every signal array from raw files. The new work selects four sites with calibration shape margins and tests their later pulse identity accuracy. Both reports use the same patients.

• usesNeighboring stimulation pairs reduced offline SEEG prediction error in five patients, but site order confounded spatial transfer

  We reused its five fixed recording pairs, trial signals, event metadata, source selection rules, raw retrieval code, and extraction code. We independently rebuilt all input arrays and used a new held-site voltage prediction test.

• usesCalibration shape selection gave higher mean site accuracy than random four-site sets for 6 mA pulses in five SEEG sessions at a 15 mm distance limit

  We used its trial bins, source metadata, fixed recording pairs, and raw retrieval and extraction code. We rebuilt all arrays from verified source files. We added a calibration rule that selects four sites and tested it against random and energy selection.

• usesMean estimated SEEG site map alignment was 0.28 for 20 to 100 ms and 100 to 500 ms in five sessions under stable means and independent group errors

  Used its reported early and late site decoding comparison as scientific context. The current calculation measures site-map alignment. No code or predictions were imported. Both reports use the same source patients.

• usesBipolar subtraction changed site accuracy in both directions at 30–100 ms in five SEEG sessions

  Use the original contact choices and source extraction preserved through the later calibration study. Its known time-window results informed the fixed early, late, and before-pulse controls. This analysis compares recording representations under a new fixed first-four and last-four split.

• usesVoltage-distance selection gave higher mean five-site accuracy than response-strength selection in 15 bipolar SEEG tests from five patients, with amplitude retained

  Uses its early-response site-identification question as context. Tests a distinct five-site selection task and does not compare its accuracy values directly.

• usesMean subtraction before early bin removal gave 7.9% site accuracy with zero late SEEG inputs in five offline patient tests

  Its separate-window centering informed the scope of our question. Our test concerns mean subtraction before bin removal and gives no basis to change that earlier separate-window result.

• usesFour calibration pulses per site selected eight sites with higher mean later-pulse identification accuracy than random sets in five SEEG sessions

  Its archived 5 ms signal bins, eligible event rows, source metadata, raw-file manifest, and raw retrieval and extraction code enabled this new small-site-set test.

• usesA joint SEEG shape decoder did not show a clear mean gain over one selected pair in four patient sessions

  The single-pair shape decoder and matched pre-pulse control informed the new joint-pair comparison. Its raw extractor supplied the MEF gap-flag check. Both studies use the same source patients.

• usesFive SEEG sessions did not establish higher site accuracy for four-pulse voltage averaging than uniform-tie voting at 20–100 ms

  We used its fixed recording contacts, trial selection, source metadata, first-half per-site split, unfiltered voltage features, and Python retrieval and extraction code. We checked all archived inputs and rebuilt every feature array from the original channel files. The new test compares repeated-pulse combination rules on fixed later groups.

• usesSEEG data after a pulse gave higher site accuracy than data before a pulse for neighboring stimulation pairs in five offline sessions

  The earlier response-window and before-pulse results informed the local plan. The new test restricts candidates to neighboring sites on one shaft. The patients are shared.

• usesA fixed confidence rule did not show a clear site identification gain over matched random rejection within five SEEG sessions

  We used the signal bins, selected contacts, trial tables, source metadata, manifest, and retrieval and extraction code. We added a fixed 4/2/later split, confidence rejection rules, and an exact matched random-rejection control. A new full raw extraction reproduced all input arrays. The patients are shared, not an independent replication.

• usesLate SEEG features gave no clear accuracy gain at eight values per pair in five offline sessions

  This earlier waveform reader test gave the response window context. Our test fixes the feature count and calibration pulse count.

• usesEarly pulse calibration selected eight sites with higher later SEEG identification accuracy than random sets within site blocks in five patients

  We use its public SEEG signal bins, event rows, source hashes, fixed contact pairs, and unchanged raw retrieval and extraction code. Its early-window result motivated the tested interval. We add calibration-only selection of sets of eight sites and later-pulse comparisons.

• usesA calibrated late SEEG score gave a 1.02-point mean site accuracy gain with a 95% interval across zero in five sessions

  We used its early and late waveform-reader comparison to choose the question. The current test uses more recording pairs, four calibration pulses per site, and calibrated covariance-score combination.

• usesA fixed waveform confidence rule gave 24.9 percent mean site accuracy on accepted later pulses within five SEEG sessions

  This test uses all five selected patient signal arrays, trial tables, source metadata, and unchanged retrieval and extraction code from the earlier benchmark. Its first-half accuracy was reproduced as a computation check. The new work adds separate calibration pulses and a control for accepted site mix.