Whole-brain multivariate hemodynamic deconvolution for functional MRI with stability selection.

Area of Science:

• Neuroimaging and Computational Neuroscience
• Analysis of functional Magnetic Resonance Imaging (fMRI) data

Background:

• Conventional fMRI analysis requires accurate experimental paradigm information, which is often unavailable in resting-state, naturalistic, or clinical scenarios.
• Existing paradigm-free methods like hemodynamic deconvolution have limitations including sensitivity to regularization parameters, lack of whole-brain analysis, and absence of statistical certainty measures.

Purpose of the Study:

• To develop a novel, multivariate, paradigm-free hemodynamic deconvolution algorithm addressing limitations of current methods.
• To improve the analysis of brain activity dynamics when experimental timing information is absent or unreliable.

Main Methods:

• Introduction of multivariate sparse paradigm-free mapping (Mv-SPFM), a whole-brain algorithm incorporating spatial information via mixed-norm regularization.
• Implementation of a stability selection procedure to eliminate the need for regularization parameter tuning and provide statistical probability of neuronal events.
• Development of a multi-echo fMRI formulation (MvME-SPFM) for enhanced BOLD signal isolation and physiologically interpretable units (ΔR₂^∗), demonstrating comparable performance with single-echo data.

Main Results:

• Mv-SPFM outperforms existing state-of-the-art deconvolution techniques in spatial and temporal agreement with standard model-based analysis.
• The stability selection procedure reduces dependency on regularization parameters (λ and ρ), enhancing algorithm performance.
• The method provides reliable estimates of neuronal-related activity, quantified as ΔR₂^∗, even for individual neuronal events.

Conclusions:

• Mv-SPFM offers a robust and reliable approach for analyzing brain activity dynamics from fMRI data, particularly when paradigm information is lacking.
• The algorithm's ability to provide statistical certainty and improved spatial/temporal resolution enhances the study of brain function.
• The Mv-SPFM algorithm will be publicly available via the splora Python package.