Modulation of Implicit Sequence Learning and Neural Oscillations by Acupuncture at Baihui (GV20) and Sishencong (EX-HN1): A Preliminary Study Using the Alternating Serial Reaction Time TASK and 256-Channel EEG

Background. Acupuncture at Baihui (GV20) and Sishencong (EX-HN1) is widely used in rehabilitation, yet whether it modulates the neurocognitive processes that support learning remains contested. Implicit sequence learning is a candidate target because it proceeds largely without awareness and can be indexed both behaviourally and electrophysiologically. Before a definitive randomised trial can be justified, it must be established that an alternating serial reaction time (ASRT) task combined with a 256-channel EEG montage can be delivered longitudinally and analysed coherently. Methods. Eight healthy, right-handed adults (4 female; mean age 20.6 ± 1.4 years) completed a within-subject T0–T1 pilot. T0 was assessed before acupuncture; T1 followed a four-week course of manual acupuncture at GV20 and EX-HN1 (three sessions per week, 30 minutes of needle retention, no electrical stimulation). Task version was counterbalanced (A→B, n = 4; B→A, n = 4). Each session comprised 20 blocks of 85 ASRT trials (1,700 trials per session). Pre-specified neural features were frontal-midline theta (4–7 Hz), sensorimotor beta (13–30 Hz) and centro-parietal P3 mean amplitude (0.3–0.6 s). Inference was at the participant level using paired contrasts with 95% confidence intervals and standardised effect sizes; theta and beta formed one Holm-corrected family and P3 was secondary. Results. Of 27,200 presented trials, 72.6% (T0) and 72.8% (T1) yielded valid reaction times, and 71.8% and 70.8% yielded valid EEG. Mean session duration was 60.0 minutes. The reaction time learning effect was 27.55 ms at T0 and 40.74 ms at T1, a paired difference of 13.19 ms (95% CI 5.84 to 20.54; t7 = 4.24, p = 0.004, dz = 1.50), with no reliable change in the accuracy learning effect (Δ = 0.0086, 95% CI −0.0099 to 0.0272, p = 0.309). The theta probability difference rose from 0.711 to 0.791 dB (Δ = 0.081 dB, 95% CI 0.026 to 0.135, Holm p = 0.010) and beta from 0.380 to 0.429 dB (Δ = 0.049 dB, 95% CI 0.037 to 0.061, Holm p < 0.001); P3 changed in the same direction but imprecisely (Δ = 0.105 μV, 95% CI −0.004 to 0.215, p = 0.057). The probability × block × session interaction was non-significant for every outcome (all p ≥ 0.087). Conclusions. The task, the cleaning rules, the ROI feature extraction, the paired contrasts and the coupling models all run on one coherent dataset, and the pilot yields the variance and retention parameters needed to size a definitive trial. Because the design was uncontrolled and the sample small, the numerical differences reported here are feasibility information only: they do not demonstrate that acupuncture at GV20 and EX-HN1 modulates implicit sequence learning or its neural correlates.