Non-extensive entropy and the extraction of BOLD spatial information in event-related functional MRI.

Functional magnetic resonance imaging (fMRI) data analysis has been carried out recently in the framework of information theory, by means of the Shannon entropy. As a natural extension, a method based on the generalized Tsallis entropy was developed to the analysis event-related (ER-fMRI), where a brief stimulus is presented, followed by a long period of rest. The new technique aims for spatial localization neuronal activity due to a specific task. This method does not require a priori hypothesis of the hemodynamic response function (HRF) shape and the linear relation between BOLD responses with the presented task. Numerical simulations were performed so as to determine the optimal values of the Tsallis q parameter and the number of levels, L. In order to avoid undesirable divergences of the Tsallis entropy, only positive q values were studied. Results from simulated data (with L = 3) indicated that, for q = 0.8, the active brain areas are detected with the highest performance. Moreover, the method was tested for an in vivo experiment and demonstrated the ability to discriminate active brain regions that selectively responded to a bilateral motor task.

Shannon entropy applied to the analysis of event-related fMRI time series.

Event-related functional magnetic resonance imaging (ER-fMRI) refers to the blood oxygen level-dependent (BOLD) signal in response to a short stimulus followed by a long period of rest. These paradigms have become more popular in the last few years due to some advantages over standard block techniques. Most of the analysis of the time series generated in such exams is based on a model of specific hemodynamic response function. In this paper we propose a new method for the analysis of ER-fMRI based in a specific aspect of information theory: the entropy of a signal using the Shannon formulation, which makes no assumption about the shape of the response. The results show the ability to discriminate between activated and resting cerebral regions for motor and visual stimuli. Moreover, the results of simulated data show a more stable pattern of the method, if compared to typical algorithms, when the signal to noise ratio decreases.