Workflows

Participant Workflow

_images/workflows-1.png

(Source code, png, svg, pdf)

The general workflow for a participant models the beta series for each trial type for each BOLD file associated with the participants. Those beta series images are output for the user. if atlas_img and atlas_lut are defined, then betas within an atlas parcel are averaged together. All the parcels are correlated with each other for each trial type, resulting in a final correlation (adjacency) matrix for each trial type.

BetaSeries Workflow

Least Squares- Separate (LSS)

from nibetaseries.workflows.model import init_betaseries_wf
wf = init_betaseries_wf(
    estimator='lss',
    fir_delays=None,
    hrf_model='glover',
    high_pass=0.008,
    smoothing_kernel=0.0,
    selected_confounds=[''])
_images/workflows-2.png

(Source code, png, svg, pdf)

nistats is used for modeling using the “least squares- separate” (LSS) procedure with the option for high pass filtering and smoothing.

Finite BOLD Response- Separate (FS)

from nibetaseries.workflows.model import init_betaseries_wf
wf = init_betaseries_wf(
    estimator='lss',
    fir_delays=[0, 1, 2, 3, 4],
    hrf_model='fir',
    high_pass=0.008,
    smoothing_kernel=0.0,
    selected_confounds=[''])
_images/workflows-3.png

(Source code, png, svg, pdf)

Additionally, NiBetaSeries can be used to perform finite BOLD response- separate (FS) modeling by combining the LSS estimator with a FIR HRF model and a set of FIR delays. This model produces a 4D beta series for each condition, at each FIR delay.

Least Squares- All (LSA)

from nibetaseries.workflows.model import init_betaseries_wf
wf = init_betaseries_wf(
    estimator='lsa',
    fir_delays=None,
    hrf_model='glover',
    high_pass=0.008,
    smoothing_kernel=0.0,
    selected_confounds=[''])
_images/workflows-4.png

(Source code, png, svg, pdf)

For completeness, NiBetaSeries also implements least squares- all (LSA), where each trial is given its own regressor in a single model as opposed to LSS where there are as many models as there are trials. While computationally faster, this method is unable to provide accurate estimates if the trials are too close together.

Correlation Workflow

_images/workflows-5.png

(Source code, png, svg, pdf)

The beta series file has signal averaged across trials within a parcel defined by an atlas parcellation. After signal extraction has occurred for all parcels, the signals are all correlated with each other to generate a correlation matrix. This step is optional.