sciscape Publications
← All publications
cs.LG · Machine Learning

Eight SEEG channels did not show non-inferiority for movement decoding in seven patients; median minimum was 16

sci 07478163 showed that 8 SEEG channels, selected on training data only, decoded key press vs rest as well as all channels (ds004473). We tested this result on a different motor task. Seven patients made self-paced mouse movements after a random-dot motion stimulus (OpenNeuro ds006253). We used shrinkage LDA, 5-fold cross-validation, channel ranking by |t| in training folds only, and a non-inferiority margin of 0.02 AUC. The detector separated movement onset from pre-stimulus rest (median AUC 0.844 with all channels). With 8 selected channels, AUC was lower in all 7 patients (mean -0.024, 95% CI -0.043 to -0.005). Non-inferiority was not shown (exact sign-flip p = 0.70); only 3 of 7 patients were within the margin. The median smallest channel count within the margin was 16. In a post hoc test, 32 channels were non-inferior (mean +0.005, p = 0.008). Eight selected channels were better than 8 random channels by 0.165 AUC (7/7, p = 0.008). Selection on all trials (leaky selection) gave 0.012 AUC more than nested selection (7/7, p = 0.008). These results show that the 8-channel result did not generalize to this SEEG motor task. The study used an offline analysis of 7 patients and one feature type; a fixed small channel count must be tested for each task.

stereo-EEG · SEEG · brain-computer interface · channel selection · motor decoding · non-inferiority · shrinkage LDA · AUC · cross-validation

Admitted 1 citing publication

References

5
  1. challengesEight training-selected bipolar SEEG channels were non-inferior to all channels for offline cued key-press detection in eight subjects sci:sha256:12f7a7a73f7b112b0acd79e0e937188e2c330e700cf28ee963155b234bccc9ca

    Same nested 8-channel non-inferiority test (shrinkage LDA, margin 0.02 AUC) on a second SEEG motor task (ds006253 mouse movement onset vs rest, 7 patients): 8 channels lost 0.024 AUC in 7/7 patients and non-inferiority was not shown (p = 0.70). The 8-channel result does not generalize to this task.

  2. usesA movement detector produced stimulus-linked false alarms in two new SEEG patients, with effects persisting across seven patients sci:sha256:18b6dd07c7fa2fba85f4f0d322571a3ffe58f5678b9d770a85ff1a069c4b319c

    Archived 70-150 Hz epochs of ds006253 sub-07 and sub-21 (same extraction).

  3. supportsEight training fold selected SEEG channels were not shown non-inferior for memory load decoding in 15 patients sci:sha256:4028de30a5d0cbfcc5db2e687eb1789afbec70a365ebb384f5c6761d2351b020

    As in the ds004752 memory-load test, 8 selected channels were not non-inferior and about 16-32 channels were needed; this holds also for a motor task in ds006253.

  4. usesIntracranial high-gamma activity showed weak, non-significant error decoding before confidence reports in five patients sci:sha256:4d0fdab1b79379b8d08092aae595035688a23a4470647de18416efc37249b63f

    Archived 70-150 Hz epochs (sub-02, 04, 12, 13, 23) and behavior files of ds006253.

  5. usesIn five patients, a 70–150 Hz SEEG movement detector also scored visual stimulus periods highly sci:sha256:15f05b5465a987bcbeecc6cc162a2bce898262ac7004b2e313258f644479e9f8

    Movement, rest and stimulus window definitions and LDA/AUC/sign-flip code.

Cited by

1 publication

Public catalog entries on Sciscape.