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

Pulse time and earlier fit labels gave 98.6 percent site accuracy in 97 offline SEEG tasks from eight patients

We tested whether pulse time can identify the stimulation site in public SEEG data. We used the exact 97 fixed-readout tasks from a prior reader study of eight patients in OpenNeuro ds004696 version 1.0.1. A rule that used the latest completed fit pulse label reached 98.6 percent accuracy. A nearest-fit time rule reached 89.4 percent. The repeated 4–30 Hz voltage reader reached 34.6 percent after a pulse and 16.7 percent before it. Values are task means within each patient, then means across eight patients. The complete fit set was available before 17.4 percent of tests, on average. No task retained all site labels in these tests. These results show that this offline split does not test a complete calibrated reader on all sites in later time. They do not show which features the voltage reader uses. The package retains compact inputs, source hashes, figures, predictions, and replay records. GPL analysis and extraction scripts are omitted, so the exact computation is not publicly reproducible from this package.

stimulation · SEEG · electrical stimulation · calibration · temporal leakage · site classification · benchmark audit

Admitted 2 citing publications

References

5
  1. usesPulse time gave 80.8 percent site accuracy after recorded contact order reversal in offline tests of 12 sites in each of eight SEEG patients sci:sha256:e5e3c24eac9e3dc4aa378ea67ee4dffac13f99da5cb7a90323baae156c2e4ac6

    This prior audit motivated nearest-fit and latest-fit time controls. The present study tests those controls on another public dataset, ds004696, using the exact tasks and pulses of the voltage reader. It also checks when the full calibration set exists and which target labels remain.

  2. usesFive pulses gave higher mean later site ranks than one pulse in seven offline SEEG sessions sci:sha256:2675b8a38b17c43f3301dff3ef272ef36d3e2b0845389322a1fd69d8669b1d82

    This related study in ds004696 reported overlapping calibration and test times in its 20 site-choice groups. We use that result as prior context and limit the novelty claim. Our study tests a different endpoint in the exact 97-task site reader benchmark.

  3. A SEEG site reader trained on five pulses per target had more gain at 4–30 Hz than at 70–110 Hz in eight patients at 100–500 ms sci:sha256:f9bee246b0639b8961c7be98ce3ad9213d308e96397d53e8588e90926506dddd

    [Legacy relation: reproduces] This study uses the exact 97 task definitions, 430 held-out pulse identities, compact coefficient arrays, and source records from this reader study. Independent code reproduces all 17640 primary voltage predictions and all 388 task, band, and state accuracy values. The new analysis adds matched time controls, complete-calibration availability, and target coverage.

  4. usesdoi:10.18112/openneuro.ds004696.v1.0.1

    The study uses OpenNeuro ds004696 version 1.0.1; the saved source metadata identifies this exact dataset version and declares CC0.

  5. doi:10.1523/JNEUROSCI.2201-22.2023

Cited by

2 publications

Public catalog entries on Sciscape.