tedana.metrics.external.calculate_max_rp_corr

calculate_max_rp_corr(*, mixing: ndarray, regressors: ndarray, seed: int | None = 0) → ndarray

Calculate the maximum regressor-correlation (max_rp_corr) for each component.

Computes the mean, over 1000 random 90%-subsamples of timepoints, of the maximum absolute Pearson correlation between each component time series and a 6*N-regressor model built from N input regressors (raw N parameters, their derivatives, and both sets time-shifted ±1 TR). Correlations are computed for the raw time series and their element-wise squares, giving 12*N total comparisons per split.

Parameters:

• mixing ((T x C) array_like) – ICA mixing matrix where T is time points and C is components.

• regressors ((T x N) array_like) – Regressor time series with T timepoints and N columns.

• seed (int or None, optional) – Seed for the random subsampling. The default is 0 for reproducible results. Set to None or -1 for stochastic sampling.

Returns:

max_rp_corr ((C,) numpy.ndarray) – Maximum regressor-correlation score for each component. Values are in [0, 1].

Notes

This metric is taken from ICA-AROMA [1], but the corresponding classification step cannot be reproduced within tedana, as ICA-AROMA combines this metric with its “edge fraction” metric, which has not been implemented in tedana. As such, any decision-tree thresholds based on this metric must be considered experimental.

The expanded regressor model has 6 * N columns before squared comparisons.

Raises:

ValueError – If regressors is not 2-D or its row count does not match mixing.

References

[1]

Raimon HR Pruim, Maarten Mennes, Daan van Rooij, Alberto Llera, Jan K Buitelaar, and Christian F Beckmann. Ica-aroma: a robust ica-based strategy for removing motion artifacts from fmri data. Neuroimage, 112:267–277, 2015. URL: https://doi.org/10.1016/j.neuroimage.2015.02.064, doi:10.1016/j.neuroimage.2015.02.064.