EEG analysis software built for the decisions that change the result

ICA review with components you actually label, channel scoring you confirm, epoching with rejection you can see, and a time-frequency view that computes once and derives seven views from it.

An EEG result is decided long before the classifier runs. It is decided when someone chooses which independent components were ocular, which channels were dead rather than merely noisy, and where the trial window began relative to the cue. Software that hides those choices behind a default is not saving you time; it is deciding your paper for you.

BCILattice puts each of them in front of you. It imports EDF, BDF, GDF, FIF, VHDR, SET, CNT and tabular EEG files, reads the sampling rate and montage from the file itself, and runs everything on your own machine.

ICA review, not ICA autopilot

ICA decomposes the recording into independent components; the part that matters is deciding which ones are artifact. The ICA Review step shows the components, lets you label the artifactual ones, and reconstructs the recording without them — as a new named stage, leaving the original untouched.

Because reconstruction writes a stage rather than overwriting one, an ICA decision is reversible. Change your mind about a component and re-run; the downstream steps read the new stage and the old one is still there to compare against.

Channel quality and channel selection are two different jobs

Channel Quality scores every channel and flags dead or noisy sensors, and you confirm the drops. That is a data-integrity step: those channels are not carrying usable signal for anyone.

Channels Selection is a modelling step further down the suite. It ranks channels by how much class information they carry — 20 criteria are available — and keeps the top ones. Conflating the two is how a channel that was merely uninformative for one contrast gets thrown away for the whole study.

Epoching, baseline correction and rejection

Epoching cuts the continuous recording into labelled trials around events, with baseline correction and trial rejection in the same step. Rejection is visible: you see what is being dropped and why before the stage is written.

For multi-subject work, set the epoching parameters against one recording in the View tab, then apply exactly those settings across the folder from the Apply tab. Both call the same code path, so the batch cannot drift from the preview.

Topographies, cortical surface and time-frequency

  • Activation Maps

    2D scalp topographies and a 3D cortical surface rendered from epoched data. View-only — it renders figures rather than writing a stage, so it never silently alters the pipeline.

  • Time-Frequency

    ERSP and ITC decomposition. One computation is cached and seven derived views are read from it, so switching view does not re-run the transform.

  • ERD/ERS

    Event-related desynchronisation and synchronisation are available as an extractable feature family, not only as a picture — so a motor imagery contrast can go straight into the design matrix.

  • 10-20 region features

    Region band power over frontal, central, parietal, occipital and temporal ROIs, derived from 10-20 electrode names in your montage.

Features that keep their names

Every extracted column is named. When the ranked feature list says relative beta power at C3 is doing the work, that is a sentence you can put in a paper and a reviewer can argue with. When it says feature 412, it is not.

Seventy-one families are available across time, frequency, time-frequency, spatial and connectivity domains — including CSP, covariance and Riemannian tangent space on the spatial side, and PLV, PLI, wPLI and coherence on the connectivity side.

Questions

Which EEG file formats can it read?

EDF, BDF, GDF, FIF, VHDR (BrainVision), SET, CNT, and tabular TXT, CSV and XLSX. A folder of recordings can be imported in one action for multi-subject studies.

Is it an EEGLAB or MNE replacement?

It overlaps with both and reads the files both produce, so it can take over from any point rather than replacing a working pipeline wholesale. The difference is scope: analysis, model training, results statistics and publication are one project here rather than four tools.

Does EEG analysis need a GPU?

No. The analysis suite runs on CPU. A GPU only matters for training deep architectures later in the ML Suite, and even there it is optional — classical pipelines train on CPU in seconds.

Can I see what a step did after the fact?

Yes. Each step records the parameters it ran with alongside the stage it wrote, so the provenance of a result is part of the project rather than something reconstructed from memory.

Read the reference

Related

Run a study end to end, on your own machine

The free plan covers two studies, the full analysis suite, and 118 of the 119 ML blocks.