Version 0.4.3🔗

Enhancements🔗

  • Align MEG artifact noise subspace scoring in SEnSAI (_sensai_score_loop) to the exact sum of sixth-power singular values (np.sum(s**6)) matching MATLAB GEDAI, dramatically improving denoising fidelity on multi-dimensional sensor configurations such as planar gradiometers (Tomas Ros).

  • Broaden broadband upper threshold bound to 12.0 for non-MEG data and align detail-band bound frequency ranges with MATLAB (min_thresh = -6.0 if (lowcut <= center_freq <= 60.0) else 0.0) across AdaptiveMultibandGedai and MultibandGedai (Tomas Ros).

  • Standardize min/max threshold bound calculation between epoch-based and raw-data fit paths (Tomas Ros).

Bugs🔗

  • Fix under-penalization of multi-component artifact leakage in MEG SENSAI by replacing the mean singular value penalty with the additive sum (Tomas Ros).

  • Ensure configured wavelet low-cutoff frequency is properly respected during frequency-adaptive threshold bound calculation (Tomas Ros).

Authors🔗

On this page