The BeMoBIL Pipeline for automated analyses of multimodal mobile brain and body imaging data

Key results

1. 01

   SuggestedThe import step converts Lab Streaming Layer recordings (extensible data format, XDF) to the Brain Imaging Data Structure (BIDS) and then to EEGLAB datasets. In these datasets every stream has the EEG sample count and the same event markers, so events found in one modality can be copied to another.

   EEG is resampled with a filter-based method (pop_resample), and the other streams are interpolated (pchip) onto the EEG sample times. In the Figure 2 example, 500 Hz EEG and 90 Hz motion both end at 250 Hz. The authors state that filter-based resampling of long recordings can drift by several hundreds of milliseconds, while interpolation keeps alignment within one sample; neither claim is quantified. · Time-synchronized data collection and import; Standardizing MoBI data; Figure legends

2. 02

   SuggestedRepeated k-means clustering picks the group-level independent component (IC) cluster for a region of interest (ROI). It ranks many clustering solutions by subject coverage, ICs per subject, spread, mean residual variance, distance to the ROI and Mahalanobis distance from the median solution.

   The example used four participants from the visual discrimination study, 1000 repetitions and a posterior parietal ROI (MNI [0,-48,39]). The top-ranked solutions were similar and the top 2 were identical. The authors state that 100 repetitions may be enough for visual cortex, while deeper ROIs such as the retrosplenial complex may need several thousand. · Robust group-level source analysis for regions…; Figure legends

3. 03

   SuggestedBlink, saccade and step events can be extracted from independent component time series when no eye tracker or motion capture was recorded. The IC-based blink detector agreed with the pupil-based eye-tracking detector around 80% of the time.

   The agreement comes from the authors' tests in one experiment with strong eye movements; participant and event counts are not given. Strong vertical eye movements can look like blinks in the vertical IC, so the authors say saccade output is not always reliable and let users switch it off. Detection uses moving-median smoothing (0.08 s default) and findpeaks. · IC-based blink and saccade event detection; IC-based gait-related event detection

4. 04

   ClaimedBefore independent component analysis (ICA), the pipeline uses the built-in sample rejection of Adaptive Mixture ICA (AMICA) rather than a separate time-domain cleaning step. Flagged samples are ignored when the spatial filters are fitted but stay in the data.

   Samples whose log-likelihood falls more than a set number of standard deviations below the mean (3 recommended) are removed in repeated passes (5 to 10 recommended). In the authors' tests this matched or beat their own epoch-based cleaning, but those results are in supplementary material not in the text. Stricter rejection may lower event-related potential (ERP) signal-to-noise despite a better decomposition. · Independent component analysis; Time-domain cleaning based on epoch rejection

5. 05

   ClaimedICA is fitted on a copy of the data high-pass filtered at a true 1.25 Hz cutoff. The decomposition is then copied back to the unfiltered data, so the filter only affects the decomposition.

   The recommended filter order is 1650 (0.5 Hz transition bandwidth) with a 1.5 Hz passband edge, using a zero-phase Hamming-window finite impulse response (FIR) filter; 2 Hz or higher is suggested for strong movement. The choice rests on the authors' earlier filter comparison, not on new data here. A final 0.2 Hz high-pass is recommended for the output dataset. · Independent component analysis; Original data reconstruction and independent…