Optimization of spatial filters for extraction of ROI time series based on the cross-talk function (CTF) or source reconstruction of spatial patterns (REC). Work in progress!

TODO

The toolbox is designed to be compatible with MNE-Python as much as possible, and optimization only requires:

• fwd: mne.Forward - the description of the forward model
• label: mne.Label - the description of the region of interest (parcel/label)

There are several parameters of the optimization that need to be specified by the user:

• lambda_ - this parameter allows fine-tuning the cross-talk function according to the demands of the analysis:
  • lambda_ = 0 prioritizes the CTF ratio, leading to better localization of sources potential contributing to the extracted signal
  • lambda_ = 1 prioritizes the CTF homogeneity, allowing to force contributions from the sources within the ROI to be more similar to a pre-specified template (see below)
  • values between 0 and 1 lead to a compromise between ratio and homogeneity
  • lambda = auto (experimental) - automatically selects the value of lambda_ to obtain a fraction (controlled with threshold) of the maximal CTF ratio or homogeneity (controlled with criteria):
    • threshold - a number between 0 and 1 controlling the fraction
    • criteria - use either ratio (criteria="ratio") or homogeneity (criteria="homogeneity") for automatic suggestion of lambda_
• template - this parameter allows specifying the desired contributions of sources within ROIs, the following options are available:
  • mean - equal values for all sources (homogeneous contribution)
  • custom templates (e.g., gaussian) may be provided directly as an array of weights for all sources within the ROI

Obtain a spatial filter that optimizes CTF properties:

from roiextract import ctf_optimize_label

sf = ctf_optimize_label(fwd, label, template, lambda_)

sf, props = ctf_optimize_label(fwd, label, template, lambda_, quantify=True)

sf = ctf_optimize_label(fwd, label, template, lambda_='auto', threshold=0.95)

Plot the filter as a topomap:

sf.plot(info)

Apply it to the data to obtain the time course of activity in the ROI/label:

label_tc = sf.apply(data)

Estimate the CTF for the filter:

ctf = sf.get_ctf_fwd(fwd)  # ctf is an instance of mne.SourceEstimate

Plot the CTF on the brain surface:

ctf.plot()