Maxwell filtering¶

About Maxwell filtering¶

The principles behind Maxwell filtering are covered in relevant references [1] [2] as well as the Elekta MaxFilter manual; see those materials for basic principles and algorithm descriptions.

In mne-python Maxwell filtering of raw data can be done using the mne.preprocessing.maxwell_filter() function.

Our Maxwell filtering algorithm currently provides multiple features, including:

• Bad channel reconstruction
• Cross-talk cancellation
• Fine calibration correction
• tSSS
• Coordinate frame translation
• Regularization of internal components using information theory
• Raw movement compensation (using head positions estimated by MaxFilter)
• cHPI subtraction (see mne.chpi.filter_chpi())
• Handling of 3D (in addition to 1D) fine calibration files
• Epoch-based movement compensation as described in [1] through mne.epochs.average_movements()
• Experimental processing of data from (un-compensated) non-Elekta systems

Movement compensation¶

When subject head movements are recorded continuously using continuous HPI (cHPI) and subjects are expected to move during the recording (e.g., when recording data in children), movement compensation can be performed to correct for head movements. Movement compensation can be performed two ways:

1. Raw movement compensation: mne.preprocessing.maxwell_filter() using the pos argument.
2. Evoked movement compensation: mne.epochs.average_movements().

Each of these requires time-varying estimates of head positions, which can currently be obtained from MaxFilter using the -headpos and -hp arguments (see the MaxFilter manual for details). The resulting MaxFilter-style head position information can be read using mne.chpi.read_head_pos() and passed to mne-python’s movement compensation algorithms.

References¶