Seven-patient SEEG replication found no clear depth effect on movement decoding, but could not establish equivalence

We tested whether deep (tip-side) or superficial (entry-side) shaft halves carry more movement information in a second SEEG center and task. We analyzed archived 70–150 Hz high-gamma window means from 7 patients performing mouse movements in a motion-direction task (187–300 trials; 49–96 same-shaft bipolar pairs). Ledoit–Wolf shrinkage LDA used a time-ordered 60/40 train/test split. The mean superficial-minus-deep test AUC was +0.036 (4/7 patients positive; exact two-sided sign-flip p = 0.56; subject-bootstrap 95% CI [-0.038, +0.129]). The prespecified replication rule required p ≥ 0.05 and the full 95% CI within [-0.05, +0.05]; it was not met. Thus these data show no significant depth difference, but cannot establish that the halves are equivalent. The superficial half matched the median equal-size random half (difference -0.000, p = 1.0), while each half and random halves lost about 0.037 AUC relative to all pairs. Patient effects varied: sub-09 favored the superficial half (AUC 0.948 vs 0.666), while sub-04 and sub-23 favored the deep half by 0.07–0.08. Single-pair AUC rose from 0.529 to 0.552 across five tip-to-entry bins; exploratory within-patient Spearman rho was +0.11 (4/7 positive, p = 0.13). Depth was contact number, assumed to place contact 1 at the tip; this was not checked with imaging. The study used one offline time split per patient, one session, high gamma only, and clinical epilepsy implants with few pairs near motor cortex. An implant plan cannot assume which half is better without patient-specific data.

SEEG · brain-computer interface · movement decoding · high gamma · depth · ROC AUC · replication