Threshold-Free Network-Based Statistics in Neuroscience

TFNBS

TFNBS Toolbox is a Python package for computation and generation of network-based statistics for neuroscience data (i.e. fMRI, EEG data). The core concept is based on eliminating the use of a hardcoded threshold using threshold-free cluster enhancement (TFCE) scores to assess statistical significance. It works on the principle of networks, where TFCE statistical values are computed across a range of thresholds over n cycles of permutations to uncover possible significance in the data. Our implementation of TFNBS follows efficient computing and allows for computations to be performed on parallel cores therefore massively reducing computation time and resources.

Installation

TFNOS toolbox can be installed using:

    !pip install tfnbs

Documentation

For more information on TFNBS's features, modules and usage, please refer to the official documentation.

Examples of usage on fMRI and EEG data are avaialble in notebooks and data.

Permutation p-values (compute_p_val)

The main entry point for permutation-based inference is tfnbs.pairwise_stats.compute_p_val.

Supported method values:

• tstat: raw t-statistics (max-stat correction)
• tfnbs: threshold-free NBS / TFCE-style enhancement
• nbs: classical NBS with fixed threshold (nbs_stat="extent" or "intensity")
• cnbs: constrained NBS (requires net_labels)
• ni_tfnbs: network-informed TFNBS (requires net_labels)
• fbc_tfnbs: functional-block clustering TFNBS (requires net_labels)

Minimal example (two-sample):

from tfnbs.pairwise_stats import compute_p_val
from tfnbs.utils import fisher_r_to_z

# group1, group2: arrays of shape (n_subjects, N, N), symmetric, diagonal=0
group1_z = fisher_r_to_z(group1)
group2_z = fisher_r_to_z(group2)

p_vals = compute_p_val(
    group1_z,
    group2_z,
    n_permutations=1000,
    test_type="two-sample",
    method="tfnbs",
    use_mp=True,
)

Notes:

• Constrained methods (cnbs, ni_tfnbs, fbc_tfnbs) require net_labels: ndarray[int] of shape (N,).
• NBS uses threshold and nbs_stat; TFNBS-family uses e, h, n, and start_thres (plus min_cluster_size for fbc_tfnbs).

Synthetic method-comparison example

See examples/sim_method_comparisons.py which compares all methods via compute_p_val and saves GT/t-stat/1-p heatmaps:

python examples/sim_method_comparisons.py --all-scenarios --effect-size 0.25 --time-points 50 --n-permutations 50

Citing the toolbox

To cite the toolbox, you can use: doi and refer to the paper paper_doi

    [doi]

For further discussions or reports on bugs, please contact ashish@ireddy.ru