Multivariate semi-blind deconvolution of fMRI time series

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Estimating the haemodynamic response function (HRF) in functional magnetic resonance imaging (fMRI) is crucial for understanding neurovascular coupling in health and disease.
• Existing methods often rely on task-based fMRI and experimental paradigms, limiting their application to resting-state fMRI (rs-fMRI).
• Previous attempts to address this for rs-fMRI include two-step analyses or univariate joint estimations, which have limitations.

Purpose of the Study:

• To develop a novel, paradigm-free method for whole-brain HRF estimation using rs-fMRI data.
• To investigate the impact of stroke and normal aging on neurovascular coupling by analyzing HRF characteristics.
• To assess the predictive power of HRF features for individual age.

Main Methods:

• A multivariate semi-blind deconvolution approach was formulated, representing neural activity as sparse spatial maps combined with piece-wise constant temporal atoms.
• An haemodynamic parcellation was introduced, incorporating unknown HRF dilation parameters within each parcel.
• A fast alternating minimization algorithm was developed and validated on synthetic and real rs-fMRI data.

Main Results:

• The framework was applied to the UK Biobank dataset, discriminating haemodynamic territories between stroke patients and healthy controls.
• Analysis of normal aging revealed that stroke and aging induce longer haemodynamic delays in specific brain regions (e.g., Willis polygon, occipital, temporal, and frontal cortices).
• A supervised classification task demonstrated that these HRF delay features could predict an individual's age with 74% accuracy.

Conclusions:

• The developed method enables robust HRF estimation from rs-fMRI data without relying on experimental paradigms.
• Neurovascular coupling, as reflected by HRF delays, is altered in stroke and changes with normal aging.
• HRF temporal characteristics offer a promising, non-invasive biomarker for assessing neurovascular health and predicting age.