Random SPREAD placement did not outperform DENSE placement for movement decoding in seven SEEG patients

A prior result on OpenNeuro ds004473 reported higher decoding AUC when K bipolar SEEG pairs were spread across K shafts rather than placed contiguously on one shaft. We tested this result in an independent dataset and task: movement onset versus pre-stimulus rest in seven patients from OpenNeuro ds006253, using archived 70–150 Hz high-gamma features. The pre-registered analysis used shrinkage LDA, a time-ordered 60/40 train/test split, and 200 random draws per placement type and patient. For random placement, SPREAD minus DENSE mean AUC was +0.005 at K = 4 (3/7 patients; one-sided Holm-adjusted p = 0.55) and −0.018 at K = 8 (2/6; p = 0.69). No patient qualified for K = 16. Adjacent pairs showed more similar rest noise and movement effects than pairs on different shafts, but this redundancy did not make random dense blocks worse at K = 4 or 8. A secondary, uncorrected K = 2 analysis favored SPREAD by +0.016 AUC in all seven patients (p = 0.008); selected placement favored SPREAD by +0.083 at K = 8 (4/6; p = 0.078). These results do not support a general random-placement SPREAD benefit at K = 4–8 across these datasets and tasks. With seven patients, a small true benefit is not excluded. The study covers one task contrast, one feature, one decoder, and one split; slow temporal drift may contribute because the rest and movement windows occur at different times.

stereo-EEG · electrode placement · high-gamma activity · movement decoding · AUC · replication · shrinkage LDA