@niivue/meica

Multi-echo ICA denoising (a Rust port of tedana) as WebAssembly, from meica-rs. runMeica outputs match the native meica CLI byte for byte (prepareMoco results depend on the niimath and mindgrab builds).

import { loadMeica, runMeica } from "@niivue/meica";
const api = await loadMeica();
const { summary, files } = runMeica({ api, echoes, tes: [15.2, 35.32, 55.42], options: { tree: "tedana_orig" } });

• loadMeica({ threads, baseUrl }) loads single/ (browser or Node 22+). The package ships only dist/single; when the page is cross-origin isolated and threads > 1 (default: up to 8 cores), it first tries threaded/, which only a self-built baseUrl serves, and falls back to single/. baseUrl points at a copy of dist/ when a bundler does not serve the package's files.
• runMeica({ api, echoes, tes, options, mask, onProgress }): echoes are { name, bytes } NIfTI files in echo order; options use tedana's flag names (tree, tedpca, seed, maxit, maxrestart, prefix, dummy_scans, and external, the regressors TSV's text); mask is optional NIfTI bytes (tedana --mask). Returns { summary, files }. Run it in a Web Worker: it blocks for seconds.
• prepareMoco({ api, echoes, run, mask, options }) is meica --moco in the browser: run(argv, { files }) runs one niimath or brainchop-mindgrab command line over in-memory files, as fastprep's tool drivers do. TR comes from echo 1's header, as in the CLI. Returns { echoes, mask, confounds, options } for runMeica.

Memory may grow to 4 GiB; measured peaks are in the repository's DEVELOPMENT.md. LGPL-2.1-only, as tedana; see LICENSE and NOTICE (attributions, including the msun notice).

Build from a checkout: npm run build (see the repository README for toolchains).