How to Support Multi-Modal Data in EEGUnity

EEGUnity supports unified channel naming and processing across multiple signal types.

Supported channel prefixes include:

• eeg for electroencephalography

• eog for electrooculography

• emg for electromyography

• meg for magnetoencephalography

• ecg for electrocardiography

• stim for stimulation/event channels

• bio for unmatched or generic biological channels

• misc for continuous label channels (for example misc:reaction_time)

EEGUnity also accepts explicit MNE channel type strings in locator entries (for example seeg:LA1, ecog:G1, dbs:DBS1, fnirs_od:S1_D1_760). Legacy uppercase prefixes (for example EEG, EOG, STIM, Unknown) are still accepted for backward compatibility.

Recommended Workflow

1. Load dataset metadata with UnifiedDataset.

2. Filter to completed records.

3. Format channel names into <Type>:<Name>.

4. Apply modality-specific processing with batch_process.

from eegunity import UnifiedDataset

ud = UnifiedDataset(dataset_path="path/to/dataset", domain_tag="my_dataset")
ud.eeg_batch.sample_filter(completeness_check="Completed")
ud.eeg_batch.format_channel_names()

About misc: Label Channels

misc: channels are useful when a dataset includes continuous labels (for example reaction time or score trajectories) that should stay aligned with EEG samples.

When resampling data, prefer EEGUnity helpers that preserve label semantics:

• eegunity.utils.label_channel.resample_raw_with_labels

• Methods that already call this helper internally, such as epoching and resampling paths in eeg_batch

Custom Processing for Multi-Modal Workloads

For full control, use:

ud.eeg_batch.batch_process(...)

In newer versions, batch_process supports execution_mode:

• execution_mode='thread' for I/O-heavy tasks

• execution_mode='process' for CPU-heavy tasks

• execution_mode=None for strict sequential execution

Notes

• Channel naming consistency is the foundation for correct modality handling.

• If your dataset has custom channel conventions, format and validate the locator first.

• For dataset-specific metadata injection, see the kernel tutorial.