Improved decoding of neural activity from fMRI signals using non-separable spatiotemporal deconvolutions

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Biophysics

Background:

• Functional Magnetic Resonance Imaging (fMRI) is a key tool for investigating neural activity.
• Standard fMRI analyses often assume the hemodynamic response function (HRF) is space-time separable.
• Empirical evidence suggests HRF dynamics are more complex than separable models allow, but the impact on information retrieval is unclear.

Purpose of the Study:

• To determine if spatiotemporal variability in HRF, not captured by separable models, contains information about neural signals.
• To assess the impact of the space-time separability assumption on decoding neural activity from fMRI data.

Main Methods:

• Compared separable and non-separable spatiotemporal deconvolution methods.
• Used simultaneously recorded intracranial neural activity and fMRI data.
• Predicted neural activity from fMRI voxel time series around intracranial electrodes.

Main Results:

• Abandoning the space-time separability assumption significantly improved the accuracy of decoding neural signals from fMRI data.
• Non-separable HRF models captured additional information about neural activity compared to separable models.

Conclusions:

• The assumption of space-time separability in HRF models may lead to a loss of information about neural signals in fMRI.
• Non-separable HRF models offer a more accurate approach for decoding neural activity from fMRI data.
• Findings have implications for refining classical fMRI analysis techniques.