Meeting Banner
Abstract #3261

A Comparison of Parametric and Non-Parametric Blind Hemodynamic Deconvolution Methods for fMRI

Karthik Ramakrishnan Sreenivasan1, Martin Havlicek2, Gopikrishna Deshpande, 3

1 AU MRI Research Center, Department of Electrical and Computer Engineering, Auburn University, Auburn, AL, United States; 2Department of Cognitive Neuroscience, Maastricht University, Maastricht, Limburg , Netherlands; 3Department of Psychology, Auburn University, Auburn, AL, United States

In this study we present a method which uses non-parametric blind deconvolution based on homomorphic filtering to investigate the over fitting problem of existing parametric methods. We compare our method to the performance of cubature Kalman filter (CKF)-based parametric approach. Simulations were performed with both methods and estimated neuronal responses were obtained. Correlation between the simulated and estimated neuronal responses indicated that in both cases (CKF-based and homomorphic deconvolution) the temporal neuronal events were correctly estimated, indicating that parametric methods such as CKF-based approaches are not susceptible to over fitting.

Abstract PDF Presentation Video

Keywords

activity added additional advantageous amplitude applications approaches approximate aspect assumed assumptions auburn biophysical blind blood bold brain called canonical causal cognitive computer confirm connectivity considered convolution convolved correctly correlation cutoff deconvolution deconvolve dependent directly domain dynamic electrical emerging employed engineering equal especially even events excellent existing expanded expectation field filter filtering fitting frequency function functional gave generated give good grid highly homomorphic hypothetical importantly impulses input inputs investigate latent length like linear magazine make many martin maximization measure measured mimic model models nature neuronal noise nonparametric oxidation parameterized parametric particular performance plot problem process processing psychology recover representation resolution respectively response responses resultant sampled search seem separation series simulated since sources subsequent susceptible temporal though threshold transform unclear unknown valid variability variables varied visible wherein whether