Reference-free BSS-CCA

BSS-CCA separates channel data using canonical correlation with a lagged copy of the signal. Components with low lagged correlation can be removed as a muscle-artifact strategy without a reference channel [1][2][3].

Usage

from mne_denoise.bss_cca import BSSCCA

model = BSSCCA(
    sfreq=250.0,
    lag_samples=1,
    n_remove=2,
)
clean = model.fit_transform(data)  # data: (n_channels, n_times)

Key points

• Input is channel-first 2-D data or MNE Raw, Epochs, or Evoked where supported; fit learns operators and transform reuses them.

• Choose n_remove or use rho_threshold to determine the removed components. reject chooses whether the low- or high-correlation end of the canonical-correlation spectrum is treated as artifactual.

• lag_samples or lag_seconds and preprocessing affect the canonical-correlation ordering.

• Segmented operation fits one operator per block; overlap controls block overlap and diagnostics report the fitted block operators.

The ordering is a signal-model assumption, not a universal artifact label. Check the result against controls and the signal of interest.

References

---

[1]

Wim De Clercq, Anneleen Vergult, Bart Vanrumste, Wim Van Paesschen, and Sabine Van Huffel. Canonical correlation analysis applied to remove muscle artifacts from the electroencephalogram. IEEE Transactions on Biomedical Engineering, 53(12):2583–2587, 2006. doi:10.1109/TBME.2006.879459.

[2]

Anneleen Vergult, Wim De Clercq, André Palmini, Bart Vanrumste, Patrick Dupont, Sabine Van Huffel, and Wim Van Paesschen. Improving the interpretation of ictal scalp eeg: bss–cca algorithm for muscle artifact removal. Epilepsia, 48(5):950–958, 2007. doi:10.1111/j.1528-1167.2007.01031.x.

[3]

Harold Hotelling. Relations between two sets of variates. Biometrika, 28(3–4):321–377, 1936. doi:10.1093/biomet/28.3-4.321.