Nonparametric hemodynamic deconvolution of FMRI using homomorphic filtering.

Area of Science:

• Neuroimaging
• Biophysics
• Signal Processing

Background:

• Functional magnetic resonance imaging (fMRI) measures neural activity indirectly via the hemodynamic response function (HRF).
• Variability in the HRF, potentially from non-neural sources, limits the accuracy of fMRI in representing true neural activity.
• Blind deconvolution is necessary to isolate neural signals, but existing parametric models may suffer from overfitting.

Purpose of the Study:

• To develop and evaluate a nonparametric blind deconvolution method for fMRI signal analysis.
• To compare the performance of the nonparametric method against a state-of-the-art parametric approach.
• To investigate the potential for overfitting in existing highly parameterized fMRI deconvolution models.

Main Methods:

• A nonparametric deconvolution technique utilizing homomorphic filtering was developed.
• The nonparametric method was compared to a parametric model employing a biophysical hemodynamic model and Cubature Kalman Filter/Smoother.
• Both methods were tested using simulations and experimental fMRI data from the visual cortex.

Main Results:

• The nonparametric deconvolution method successfully estimated the latent neuronal response from fMRI data.
• Results from the nonparametric and parametric methods showed high correlation in the visual cortex.
• Simulations confirmed that both methods effectively recovered the ground truth of the simulated latent neural response.

Conclusions:

• The nonparametric deconvolution approach provides a viable alternative for analyzing fMRI data.
• The high agreement between parametric and nonparametric methods suggests that overfitting concerns in complex fMRI models might be overstated.
• This work contributes to more accurate estimation of neural activity from fMRI signals.