Spatial-temporal modelling of fMRI data through spatially regularized mixture of hidden process models

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Biophysics

Background:

• Functional magnetic resonance imaging (fMRI) analysis traditionally focuses on spatial patterns of neural activation.
• Detecting robust spatio-temporal constraints in fMRI data is crucial for understanding brain function.
• Existing methods often lack the ability to model the complex temporal dynamics and spatial influences inherent in fMRI signals.

Purpose of the Study:

• To present a novel mixture-based method for spatio-temporal modeling of fMRI data.
• To infer overlapping cognitive processes using the Hidden Process Model (HPM) framework with parametric Haemodynamic Response Function (HRF) estimation.
• To identify and characterize spatio-temporal patterns of neural activation, contrasting with conventional spatial-only approaches.

Main Methods:

• Developed a mixture-based model assuming fMRI time series are a probabilistic superposition of spatio-temporal prototypes.
• Employed the Hidden Process Model (HPM) framework with a parametric HRF for temporal modeling.
• Utilized spatial priors to define regions of influence for each prototype, independent of voxel count.

Main Results:

• Validated the model using synthetic and real fMRI data from a rapid event-related visual recognition experiment.
• Demonstrated that frontal regions are less homogeneous than occipitotemporal regions, requiring multiple HPM prototypes.
• Successfully disentangled perceptual judgment and motor response processes in frontal regions, highlighting spatio-temporal heterogeneity.

Conclusions:

• The proposed model offers a principled and computationally efficient approach to identify spatio-temporal activation patterns in fMRI data.
• Frontal regions exhibit significant spatio-temporal heterogeneity, linked to the dynamic localization of distinct cognitive processes.
• The method effectively distinguishes overlapping cognitive processes even in rapid event-related designs.