Background and objective: Precise estimation of neuronal activity from neuroimaging data is one of the central challenges of the application of noninvasive neuroimaging methods. One of the widely used methods for studying brain activity is functional magnetic resonance imaging, which is a neuroimaging procedure that measures brain activity based on the blood oxygenation level dependent effect. The blood oxygenation level dependent signal can be modeled as a linear convolution of a hemodynamic response function with an input signal corresponding to the neuronal activity. Estimating such input signals is a complicated problem.
Methods: We present a software tool for estimation of brain neuronal activity, which uses a combination of Wiener filtering with deconvolution methods, including the least absolute shrinkage and selection operator, the ordinary least squares method, and the Dantzig selector. The latter two are equipped with both established selection criteria (Akaike and Bayesian information criterion) as well as newly developed mixture criteria for selection of activations.
Results: The software tool was tested on two types of data: measurements during basic visual experiments and during complex naturalistic audiovisual stimulation (watching a movie segment). During testing the software showed reasonable results, with the mixture criteria performing well for temporally extended activations.
Conclusions: The presented software tool can be used for estimation, visualization, and analysis of brain neuronal activity from functional magnetic resonance imaging blood oxygenation level dependent measurements. The implemented methods provide valid results not only in the sparse activity scenario studied previously but also for temporally extended activations.
Keywords: Deconvolution methods; Functional magnetic resonance imaging; Hemodynamic response; Neuronal activity estimation; Wiener filtering.
Copyright © 2017 Elsevier B.V. All rights reserved.