HMM: Plotting fMRI Networks

We normally plot networks from an HMM trained on fMRI directly from the dual-estimated means/covariances. This tutorial covers:

1. Load means/covariances

2. Spatial maps

3. Functional connectivity networks

Note, we assume a group ICA parcellation was used and the HMM was trained on the ICA time courses.

Load means/covariances

First we need to the dual estimated means and covariances.

import numpy as np

means = np.load("results/dual_estimates/means.npy")
covs = np.load("results/dual_estimates/covs.npy")
print(means.shape)
print(covs.shape)

means is a (subjects, states, channels) array and covariances is a (subjects, states, channels, channels) array.

Spatial maps

Volumetric parcellation

The spatial activity maps correspond to the means, or if we did not learn a mean, the diagonal of the covs. How we plot the spatial maps depends on how the data was preprocessed. If we used a volumetric parcellation, then we can plot the spatial maps with

from osl_dynamics.analysis import power

# Calculate a group average
group_mean = np.mean(means, axis=0)

# Plot
fig, ax = power.save(
    means,
    mask_file="MNI152_T1_2mm_brain.nii.gz",
    parcellation_file="melodic_IC.nii.gz",  # this should be the group-ICA spatial maps from FSL
    plot_kwargs={"views": ["lateral", "medial"]},
)

Surface parcellation

Alternatively, if we used a surface parcellation, we can use workbench to plot the spatial maps

 # Save cifti containing the state maps
 power.independent_components_to_surface_maps(
     ica_spatial_maps="melodic_IC.dscalar.nii",
     ic_values=group_mean,
     output_file="results/inf_params/means.dscalar.nii",
 )

# Plot
workbench.setup("/path/to/workbench/bin_macosxub")
workbench.render(
    img=f"{inf_params_dir}/means.dscalar.nii",
    gui=False,
    save_dir="tmp",
    image_name="plots/means_.png",
    input_is_cifti=True,
)

# Delete temporary directory
os.system("rm -r tmp")

You can download workbench here.

If you did not learn the mean, then replace means above with the diagonal of the covariances.

Functional Connectivity Networks

The off-diagonal elements in the covariances corresponds to the functional connectivity. In osl-dynamics, we can only plot these if we used a volumetric parcellation with

from osl_dynamics.analysis import connectivity

connectivity.save(
    covs,
    parcellation_file="melodic_IC.nii.gz",  # this should be the group-ICA spatial maps from FSL
    plot_kwargs={
        "edge_cmap": "Reds",
        "display_mode": "xz",
        "annotate": False,
    },
)