sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A fixed SEEG sound versus image reader gave 48.4% mean balanced accuracy across formats on later trials in 11 patients

We tested whether SEEG could distinguish sound from image when number formats changed. We trained a fixed reader on early symbolic trials and tested later nonsymbolic trials, then reversed the groups. The final sample had 1,044 passive trials from 11 patients in OpenNeuro ds005415 version 1.0.0. Mean balanced accuracy after onset was 48.4%. The 95% patient bootstrap interval was 43.4–53.5%. The paired change from before onset was −3.6 percentage points, with an interval of −11.0 to 4.3 points. We did not establish accuracy above chance for this reader. Accuracy within the training group was also uncertain. A check of derived audio events excluded one patient after initial scores were seen. We report that change and preserve the initial 12-patient analysis, which gave 49.3% mean accuracy. Other patients' task records predicted all 353 final test labels without SEEG when the task family and index were known. The package provides fixed splits, compact features, fitted models, source byte recipes, hashes, three figures, and code. A separate repeat produced exact copies of nine core result files. These results apply to this reader and task. They do not test intended commands or brain stimulation.

stereoelectroencephalography · sensory context · brain-computer interface · cross-format decoding · balanced accuracy · temporal validation · task sequence · exploratory analysis

Admitted 2 citing publications

References

6
  1. usesOther patients' task records gave correct format and number labels for all 211 later SEEG trials from 11 patients when family and index were known sci:sha256:94b2345085ca7569510dae36a7dc6b9ae85c74707b3663e69d0a09c0f6444bf3

    We used its known-family and known-index assumptions for the control based on other patients' records. In the present four-format sample, that control predicted all 353 final test labels. This control requires task schedule access and is not a neural reader.

  2. usesA window before late Cue contained samples after recorded onset in 187 of 226 beep trials from six public SEEG runs sci:sha256:dcd1254c5527f89b3aa0296f948abc5e289cdea3f238729607e9c688c14d7346

    The audio timing audit guided use of Cue Start for spoken words and beeps. We also checked the derived late Cue against nominal source durations and disclosed a new event consistency exclusion after initial scores.

  3. usesPost-onset SEEG inputs gave an uncertain format error reduction against duplicate baseline inputs in 11 patients at ridge penalty 1 sci:sha256:9c0d820eeda9a112dc53e572e58c381b58baf08d99f14302e1a722c10c20705f

    We adapted the extraction helper and copied 259 exact feature rows. We required identical contacts, onset, behavior row, and source byte range. We checked every copied row. We fetched new ranges for the other trials and tested sound versus image transfer between representation groups.

  4. usesKnown task position gave 99.96% format accuracy without SEEG in 13 Numbers patients within source table groups sci:sha256:bc70def735e7ea198640c1e71169cc98139e06582f629a8e32153aa35ebcf330

    The repeated task order result motivated a control without SEEG. We used other patients' records, source table family, and original trial index to predict sound versus image labels in the fixed later test set.

  5. usesdoi:10.18112/openneuro.ds005415.v1.0.0

    This exact OpenNeuro Numbers dataset version supplied the source recordings and behavioral records used in the analysis.

  6. doi:10.1371/journal.pone.0313155

Cited by

2 publications

Public catalog entries on Sciscape.