pySPFM 2.0.2

Sparse hemodynamic deconvolution of fMRI data with scikit-learn compatible estimators.

pySPFM

A Python package for sparse hemodynamic deconvolution of fMRI data with scikit-learn compatible estimators.

pySPFM provides estimators for Paradigm Free Mapping (PFM) and related sparse deconvolution methods for fMRI analysis. The package follows scikit-learn conventions, making it easy to integrate with existing machine learning pipelines.

Features

• scikit-learn compatible API: Use familiar fit(), transform(), fit_transform() methods
• Multiple deconvolution methods:
  • SparseDeconvolution: Sparse Paradigm Free Mapping (SPFM) using LARS or FISTA
  • LowRankPlusSparse: SPLORA algorithm separating global and neuronal signals
  • StabilitySelection: Robust feature selection via bootstrap resampling
• Multi-echo fMRI support: Native support for multi-echo acquisitions
• Flexible regularization: Multiple lambda selection criteria (BIC, AIC, MAD, etc.)
• Command-line interface: Easy-to-use CLI for batch processing

Installation

pip install pySPFM

For development:

git clone https://github.com/Paradigm-Free-Mapping/pySPFM.git
cd pySPFM
pip install -e ".[dev,tests]"

Quick Start

Python API

from pySPFM import SparseDeconvolution
import numpy as np

# Load your fMRI data (n_timepoints, n_voxels)
X = np.random.randn(200, 1000)  # Example data

# Create and fit the model
model = SparseDeconvolution(tr=2.0, criterion='bic')
model.fit(X)

# Get the deconvolved activity-inducing signals
activity = model.coef_

# Get the fitted (reconstructed) signal
fitted = model.get_fitted_signal()

# Compute explained variance
score = model.score(X)
print(f"Explained variance: {score:.2%}")

Low-Rank + Sparse Decomposition (SPLORA)

from pySPFM import LowRankPlusSparse

model = LowRankPlusSparse(tr=2.0, eigval_threshold=0.1)
model.fit(X)

# Separate components
sparse_activity = model.coef_      # Neuronal activity
global_signal = model.low_rank_    # Global/structured component

Command-Line Interface

# Sparse deconvolution
pySPFM sparse -i data.nii.gz -m mask.nii.gz -o output --tr 2.0

# Low-rank + sparse decomposition
pySPFM lowrank -i data.nii.gz -m mask.nii.gz -o output --tr 2.0

# Stability selection
pySPFM stability -i data.nii.gz -m mask.nii.gz -o output --tr 2.0 --n-surrogates 50

Estimators

Estimator	Description	Use Case
SparseDeconvolution	Sparse PFM using LARS/FISTA	General deconvolution
LowRankPlusSparse	SPLORA algorithm	Separating global signals
StabilitySelection	Bootstrap-based selection	Robust feature detection

API Reference

All estimators follow scikit-learn conventions:

• fit(X): Fit the model to data
• transform(X): Return deconvolved signals
• fit_transform(X): Fit and transform in one step
• score(X): Return explained variance ratio
• get_params() / set_params(): Parameter access
• clone(): Create unfitted copy

Development

This project uses uv for dependency management and tox for testing.

Running tests

# Current Python version
pytest pySPFM/tests/

# All supported versions (3.10, 3.11, 3.12)
tox

# Specific version
tox -e py310

References

• Caballero-Gaudes, C., Moia, S., Panwar, P., Bandettini, P. A., & Gonzalez-Castillo, J. (2019). A deconvolution algorithm for multi-echo functional MRI: Multi-echo Sparse Paradigm Free Mapping. NeuroImage, 202, 116081–116081. https://doi.org/10.1016/j.neuroimage.2019.116081
• Caballero Gaudes, C., Petridou, N., Francis, S. T., Dryden, I. L., & Gowland, P. A. (2013). Paradigm free mapping with sparse regression automatically detects single-trial functional magnetic resonance imaging blood oxygenation level dependent responses. Human Brain Mapping. https://doi.org/10.1002/hbm.21452
• Karahanoǧlu, F. I., Caballero-Gaudes, C., Lazeyras, F., & Van De Ville, D. (2013). Total activation: FMRI deconvolution through spatio-temporal regularization. NeuroImage. https://doi.org/10.1016/j.neuroimage.2013.01.067
• Uruñuela, E., Bolton, T. A., Van De Ville, D., & Caballero-Gaudes, C. (2023). Hemodynamic Deconvolution Demystified: Sparsity-Driven Regularization at Work. Aperture Neuro, 3, 1-25. https://doi.org/10.52294/001c.87574