Skip to content

SPiN-Lab/connPFM

Repository files navigation

connPFM

Latest Version PyPI - Python Version DOI License CircleCI Documentation Status Codecov Code style: black DOI

Methodology

Mapping time-varying neural-related activity at TR resolution is a common goal between hemodynamic deconvolution methods that estimate the underlying neuronal-related activity of the BOLD signal and analysis of instantaneous co-fluctuations (CF) between brain regions. We present a new deconvolution approach, named connPFM, that combines the best of both methods. The connPFM method comprises of 3 steps:

  1. Deconvolution with stability-selection paradigm free mapping that computes the probability (area under the curve; AUC) that a neuronal-related event generates a BOLD event at each time;
  2. Selection of significant CF events based on the pairwise CF matrix computed from the AUC timecourses (CF-AUC); this selection step can be done in 2 ways:
    • Temporal: Thresholding of the root sum of squares (RSS) time-series calculated from the CF-AUC matrix,
    • Spatio-temporal selection: Thresholding of the CF-AUC matrix based on the on a threshold taking into account significant values of CF-AUC based on a null distribution
  3. Debiasing of the neuronal related activity associated with the selected deconvolved events, which show a significant CF with any other region, through ordinary least-squares regression. For illustration, we present results of connPFM in one mulitecho subject fMRI dataset. We show that connPFM improves the sensitivity to blindly detect functionally-relevant BOLD events in resting-state networks compared to deconvolution approaches that do not exploit CF information.

connPFM flowchart

Installation

    git clone https://github.com/SPiN-Lab/connPFM.git
    cd connPFM
    pip3 install -e .[all]