Answers are grounded in this paper and cite their sources.
Scalp electroencephalography (EEG) may contain gait information without improving forecasts beyond electromyography (EMG). We evaluated this increment in 163 treadmill trials from 57 healthy participants (145,885 prediction origins). Six bilateral lower-limb angles were forecast approximately 124.6 ms ahead on the released grid using two-second EEG band powers and EMG features, ridge regression, and 5 × 4 participant-disjoint nested validation. Shared-penalty fusion produced no false-discovery-rate-significant gain over EMG-only and reduced R2 for the left knee (ΔR2 = −0.0260, q = 0.0072) and left ankle (ΔR2 = −0.0390, q = 0.0222). Its descriptive mean increment was −0.0156 (95% CI −0.0312 to 0.0004). Separate penalties did not establish an advantage (mean ΔR2 = −0.0009, 95% CI −0.0123 to 0.0084). An additional C3/Cz/C4 analysis likewise did not establish mean improvement within its fixed exploratory grid (mean ΔR2 = −0.0006, 95% CI −0.0074 to 0.0058). Extending EEG penalties downward left selected models and predictions unchanged. Current-angle, speed-adjusted, and low-gamma-excluded comparisons revealed no reliable gain; EEG-only did not reliably outperform the training mean. These findings apply to the tested predictors in healthy walking. They do not establish equivalence, identify a neuroanatomical cause, or assess intention decoding in patients with weak, unreliable, or spastic activation.
No takes yet. Share an insight, caveat, or question.
pang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: