sciscape Publications
← All publications
eess.SP · Signal Processing

In 15 SEEG patients, a group map did not outperform the hand-knob rule for offline movement decoding

We compared two label-free rules for choosing 8 SEEG bipolar pairs in 15 patients from OpenNeuro ds006253. MAP8 ranked pairs with a 10 mm Gaussian group map built from the other 14 patients; HK8 chose pairs nearest a fixed MNI hand-knob point. Using archived 70–150 Hz high-gamma amplitude features, a time-ordered 60/40 train/test split, and shrinkage LDA, MAP8 did not outperform HK8: mean held-out AUC was 0.691 versus 0.712 (difference −0.020, 95% bootstrap CI −0.077 to +0.039; 4/15 positive; exact two-sided sign-flip p=0.519). The rules selected mostly different pairs (mean overlap 1.67 of 8). Each exceeded random 8-pair selection by 0.04–0.06 AUC in uncorrected tests, but neither comparison remained significant after Holm correction (p=0.059 for both). Patient-specific selection using labeled training data had higher AUC (OWN8 0.807; all pairs 0.832). The results apply to one offline, task-locked mouse-movement versus pre-stimulus-rest task; they do not establish online BCI performance. The study included 15 patients and had low power for differences below about 0.05 AUC.

SEEG · brain-computer interface · channel selection · high-gamma activity · movement decoding · shrinkage LDA · held-out AUC

Admitted 2 citing publications

References

6
  1. supportsA leave-one-patient-out fsaverage map improves eight-channel SEEG movement decoding over random selection in 10 patients sci:sha256:f90e6936051d3168570f64dfb5b6f5a09478ee74976822452a6a61b04eedb04a

    In 15 patients, the leave-one-patient-out map again gave higher AUC than random 8-pair sets (0.691 vs 0.651, 11/15, uncorrected p=0.030). The effect is smaller, and it is not significant after Holm correction (p=0.059).

  2. usesA leave-one-patient-out fsaverage map improves eight-channel SEEG movement decoding over random selection in 10 patients sci:sha256:f90e6936051d3168570f64dfb5b6f5a09478ee74976822452a6a61b04eedb04a

    We use the analysis code (auc, shrinkage LDA, map, sign-flip, Holm), the map method and 10 fsaverage electrode files from this package.

  3. usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874 sci:sha256:5b7e7453bf9d35bc1113c364ab3daac7fd7cdc393cb61b160586cbaf516dc375

    We use the archived 70-150 Hz HGA window features and provenance files of 15 ds006253 patients from this package.

  4. supportsIn 11 SEEG patients, a fixed hand-knob distance prior improved held-out movement-versus-rest AUC over random channel selection sci:sha256:55c1eac8130c36ee0c7fd6c298ba1b63f37fe5e183c6c430dab558c5f077dfc0

    In 15 patients, hand-knob nearest pairs again gave higher AUC than random 8-pair sets (0.712 vs 0.651, 11/15, uncorrected p=0.047). The effect is smaller, and it is not significant after Holm correction (p=0.059).

  5. usesIn 11 SEEG patients, a fixed hand-knob distance prior improved held-out movement-versus-rest AUC over random channel selection sci:sha256:55c1eac8130c36ee0c7fd6c298ba1b63f37fe5e183c6c430dab558c5f077dfc0

    We use the hand-knob nearest-pair selection rule of this work as the comparison rule HK8.

  6. usesdoi:10.18112/openneuro.ds006253.v1.0.3

    The study used archived 70–150 Hz HGA features and electrode coordinates from OpenNeuro ds006253. Feature provenance files record the source dataset keys and SHA-256 hashes; the five coordinates retrieved from the public S3 mirror are separately identified by URL, retrieval date, and packaged hashes.

Cited by

2 publications

Public catalog entries on Sciscape.