Mobile EEG for the study of cognitive-motor interference during swimming?

The study addresses whether mobile EEG can reliably measure brain activity, including cognitive-motor interference (CMI) markers, during a complex full-body movement—swimming. Mobile brain/body imaging (MoBI) enables studying cognition in motion, but participant and device mobility vary across systems. Prior work shows dual-tasking typically reduces P300 amplitudes and affects cognitive performance during activities like walking, cycling, skateboarding, and slacklining. The authors aim to extend mobile EEG into an aquatic environment with high participant and device mobility, validate signal quality using the N100 ERP component, and test expected CMI effects on the P300 amplitude and latency during an auditory oddball task performed while sitting (single-task) and swimming (dual-task). They additionally explore alpha/mu and beta band modulations around swimming turns to probe movement preparation and coordination.

Mobile EEG and MoBI have advanced ecological validity by enabling brain recordings during natural movement (Gramann et al., Debener et al., De Vos et al., Bleichner & Debener). Systems differ in mobility and specifications (Bateson et al.). ERP presence and single-trial noise are common quality metrics. CMI is well documented in various locomotor contexts, with dual-tasking typically leading to reduced P300 amplitudes and poorer cognitive/motor performance (Al-Yahya et al.; Plummer et al.; Leone et al.; Gramann et al.; Debener et al.; Lau et al.; Jain et al.; Enders et al.; Zink et al.; Scanlon et al.; Reiser et al.; Robles et al.; Papin et al.). The P300 (~300–600 ms) reflects attentional resource allocation and workload (Polich & Kok; Polich; Luck). Prior aquatic neuroscience in humans is scarce and typically stationary or non-EEG; underwater EEG in humans was stationary (Schneider et al.), while animal studies recorded EEG in sea animals and rodents. Movement-related oscillations include alpha/mu and beta desynchronization during movement preparation and control (Gross et al.; Palva & Palva; Maksimenko et al.).

Preregistration: Hypotheses and analysis plans were preregistered on OSF (submitted Dec 14, 2022; https://osf.io/qt4mr). Participants: N=11 right-handed, healthy swimmers (5 female, 6 male), aged 16–54 (M=32.18, SD=12.05), recruited via personal contacts. Inclusion required several years of swimming experience. Ethical approval: University of Oldenburg (Drs/EK2022/040-01); written informed consent; reimbursement 10€/h. Materials: Auditory oddball implemented in Presentation Mobile (Version 23.0) on Android smartphone (Samsung Galaxy S21 5G), delivered via waterproof in-ear headphones (IPX8, AGBTEK). EEG recorded with 28 passive sintered Ag/AgCl electrodes at 10–20 sites (F3, F4, C3, C4, P3, P4, O1, O2, F7, F8, T7, T8, P7, P8, Fz, Cz, Pz, POz, FC1, FC2, CP1, CP2, FC5, FC6, CP5, CP6, TP9, TP10); reference FCz, ground AFz; cap without chin strap (Easycap); impedances <20 kΩ using Abralyt HiCl gel. Mobile amplifier: SmartingPro (mBrainTrain) attached at back of cap below O1/O2. Sampling rate 250 Hz via SmartingPro App (Version 2.2). Head movements recorded via IMUs (accelerometers, gyroscopes, quaternions). Data synchronized with Lab Streaming Layer (LSL) and Receiva app; stored in XDF. Waterproofing: silicone swim cap over EEG cap and amplifier; medical tape sealing around head and upper spine; amplifier additionally taped. Swim equipment: front snorkel to minimize head rotation; buoy with smartphone to maintain Bluetooth connection above water and manage cable length. Tasks: Auditory oddball: 200 tones per block (70 ms duration), ISI 1–3 s random; standards 800 Hz; deviants 1000 Hz; 47–54 deviants per block. Subjects counted deviants and reported counts after each block. Motor conditions: Blocks Pre Swim (sitting), Swim (front crawl swimming), Post Swim (sitting). Swim contained two oddball runs to balance trial counts; fixed order to assess pre/post device functionality. Procedure: Conducted in a closed university indoor pool (~23°C). After EEG setup and resting measurement, Pre Swim oddball while seated. Swim: snorkel and goggles over swim cap; buoy belt with smartphone; underwater volume adjustment; 5-min acclimation swim; two ~7-min oddball swims separated by a 2-min break; counts reported mid-block and at end; then Post Swim sitting with eye artifact and resting measures followed by oddball; counts reported. EEG preprocessing (EEGLAB 2022.0; MATLAB R2022a): Data completeness check; exclude two subjects (IDs 05, 08) with missing recordings; include two subjects with partial sitting data (ID 10 Pre missing; ID 11 Post missing). Initial 1 Hz zero-phase FIR highpass filter (order 826); identify bad channels (SD > mean+3 SD), visually validate, temporarily remove. Merge blocks per subject; remove line noise via zapline-plus. Clean with ASR (threshold 70 SD). Perform extended infomax ICA on 1 s epoched data, excluding epochs with joint probability >3 SD or kurtosis >3 SD; manually reject eye and muscle components across blocks. Apply ICA weights to minimally processed raw data (0.3 Hz HP FIR, order 2750; 30 Hz LP FIR, order 110); ASR (70 SD); interpolate removed channels. ERP processing: re-reference to linked mastoids (TP9/TP10), epoch -0.2 to 0.8 s around stimulus; baseline correct (-0.2 to 0 s); reject epochs with joint probability >5 SD or kurtosis >5 SD. Turn-related time-frequency: use ICA-corrected Swim data re-referenced to common average; epoch -2.5 to 2.5 s around turn markers extracted from IMU channels; reject epochs >5 SD probability/kurtosis. ERP parametrization: N100 at Fz (100–200 ms): find minimum voltage for average ERP; define individual ±50 ms search window around peak; within single trials, locate minimum and average within ±25 ms around the peak; visual validation via topography. P300 at Pz (300–600 ms): find maximum of average ERP; define individual ±100 ms search window; within single trials, mean amplitude in ±50 ms window around peak used as measure; extract peak latency (time from stimulus onset to amplitude peak). Note: amplitude parametrization deviated from preregistration due to data characteristics. Exploratory time-frequency analysis: Downsample to 60 Hz (for consistency with prior motion-sensor synchronization work). Transform turn-locked epochs using Morlet wavelets. Region of interest: FC1, FC2, C3, Cz, C4; average across epochs, ROI, participants to obtain grand average. Normalize to sitting baseline to remove 1/f. Statistics (RStudio 2022.12.0): N100: Shapiro–Wilk tests for normality; one-sided t-tests or Wilcoxon rank tests against zero (alternative = less) within subjects and blocks; binomial tests to compare proportion of significant within-subject effects against chance level using qbinom-derived thresholds (Combrisson & Jerbi), chance level 77% for n=9, 75% for n=8. P300: Within-subject permutation tests (5,000 iterations) comparing target vs standard amplitudes per block; subsample standard trials to match target counts each iteration; p-values from empirical position within null distribution; binomial tests across subjects. Interaction: Recode Pre/Post as Sit, compare Sit vs Swim for target-standard amplitude differences; permutation analysis on differences; permutation analysis for P300 latencies in Sit vs Swim. Exploratory time-frequency: Divide grand average into 10 × 500 ms bins for alpha/mu (8–13 Hz) and beta (15–30 Hz); permutation tests per bin vs mean of all bins; two-sided with Bonferroni correction (α=0.0025).

Sample and preprocessing: Two subjects excluded due to recording issues, leaving N=9 for analysis. On average, 1.11 channels (range 0–3) marked bad and removed; ~21 ICA components retained post-artifact correction. Data retained post-preprocessing: Pre Swim: 122 standard, 47 target epochs (loss ~15.5%); Swim: 244 standard, 75 target (loss ~20.25%); Post Swim: 129 standard, 48 target (loss ~11.5%). Turn epochs: 24 (range 17–30) usable out of 25.5 (range 18–31) per participant. Noise: Single-trial prestimulus noise at Fz and Pz increased in Swim compared to Pre/Post in all included participants (categorical regressions p<0.001), but the proportion of significant increases did not exceed chance in binomial test (8/8 successes, p=0.1). Post vs Pre noise higher in two subjects (p<0.05), proportion not above chance (2/7, p=0.99). N100: Occurred reliably (p<0.05 within subjects) after both standard and target tones in all blocks (Pre, Swim, Post), with plausible topographies. However, binomial tests across subjects did not exceed chance thresholds (Pre/Post p=0.1; Swim p=0.104). P300 main effect (target>standard amplitude): Pre Swim significant in 6/8 subjects (p<0.05), not above chance (p=0.68). Swim significant in 2/9 (p<0.05), below chance (p=1). Post Swim significant in 5/8 (p<0.05), not above chance (p=0.89). Grand-average ERPs showed typical oddball morphology in sitting blocks; amplitudes reduced during Swim. Interaction (CMI): Target–standard amplitude difference larger in Sit than Swim was significant in 1/9 subjects (binomial p=1). P300 latency longer in Swim than Sit was significant in 5/9 subjects (binomial p=0.97), indicating no reliable group-level effect. Exploratory time-frequency around turns: Alpha/mu (8–13 Hz) power significantly decreased at -2500 to -2000 ms and -1500 to 1000 ms before the turn, and 0–500 ms immediately after (p<0.0025), and significantly increased 1500–2500 ms after the turn (p<0.0025). Beta (15–30 Hz) power significantly increased from approximately -2000 to -1500 ms before the turn and 1000–2500 ms after (p<0.0025), and decreased from -1000 to 500 ms around the turn (p<0.0025). These patterns suggest preparatory desynchronization and post-turn resynchronization consistent with motor control processes. Overall: Mobile EEG recordings in an aquatic environment were feasible, with functional N100 across conditions and detectable ERPs/time-frequency modulations despite increased noise and data loss during swimming. Robust post-swim signal quality was maintained.

The study demonstrates feasibility of acquiring meaningful mobile EEG during swimming by maintaining device integrity (waterproofed cap and amplifier), stable Bluetooth and audio delivery (smartphone in buoy; reduced cable slack; snorkel to minimize head rotation), and sufficient signal quality through preprocessing (ASR, ICA). Despite elevated noise and data loss in the Swim condition, ERPs including N100 were reliably observed, and P300 components were present in some participants. However, classical CMI markers (reduced P300 amplitude and increased latency during dual-tasking) did not emerge reliably across subjects—likely due to small sample size, intra/inter-individual variability in engagement and arousal, potential physical exertion effects, and prioritization of motor safety (“posture/not drowning first”) during swimming. The exploratory time-frequency results around turns revealed plausible alpha/mu and beta modulations indicative of movement preparation and coordination, suggesting mobile EEG can capture oscillatory dynamics in complex full-body motion. Relative to mobility frameworks, the setup achieved high device and participant mobility (head-mounted amplifier, smartphone-based recording, free swimming with stabilized head) with moderate system specs (passive gel electrodes, 24-bit, up to 500 Hz, ~5 h battery), adequate for the task. Passive electrodes performed comparably to active systems in mobile contexts. These findings support further development of mobile EEG for extreme, real-world environments, while emphasizing the need for larger samples, improved behavioral measures, and refined experimental control to robustly quantify CMI effects during swimming.

This pilot study extends mobile EEG into an aquatic environment, showing it is feasible to record ERPs and time-frequency dynamics during swimming with high participant and device mobility. The N100 occurred reliably across conditions, while P300 effects typical of oddball tasks and CMI were inconsistent across subjects, likely reflecting sample size limitations and dual-task prioritization. Exploratory analyses revealed meaningful alpha/mu and beta modulations around turns, suggesting sensitivity to movement preparation and coordination. Future work should employ larger samples, counterbalanced designs, improved waterproofing and head stabilization, and reliable behavioral measures (e.g., waterproof response devices, task difficulty ratings) to better characterize CMI and motor-related brain dynamics in swimming. Such advances could enable applications in sports diagnostics, training, and studies of the cognitive benefits of swimming.

Small sample size (N=11; 9 analyzed) limits generalizability and statistical power; access to pool time constrained recruitment. Prototype setup included a single modified cap used across participants, leading to electrode position variability due to differing head sizes. Fixed block order (Pre–Swim–Post) introduces potential order and habituation effects (P300 amplitude can decrease with repetition). No robust behavioral performance metric (count reports may include guessing; no button-press feasible during swimming) or subjective difficulty ratings. Physical exertion may have varied across participants despite instructions, potentially affecting cognitive engagement and ERPs. Increased noise and data loss in Swim relative to sitting conditions.