Dynamic Causal Modeling and machine learning for effective connectivity in Auditory Hallucination

Abstract

Abstract Hallucinations are an elusive phenomena that has been associated with psychotic behavior, but which has a high prevalence in healthy population. Some models of Auditory Hallucinations (AH) have been proposed in the literature, but so far empirical evidence is scarce. The current study aims to obtain measurement of effective connectivity on rs-fMRI showing discrimination of subjects with a history of AH to give empirical validation of AH models. In this paper, we consider the Dynamic Causal Modeling (DCM) for the estimation of the effective connection between brain regions on resting state functional magnetic resonance imaging (rs-fMRI). Applying DCM to rs-fMRI data is a very recent approach, that allows studies of effective connectivity on rs-fMRI data. Brain regions are selected on the basis of previous results on functional connectivity in agreement with the models proposed in the literature. The selection of the most significant connection structure is carried as a wrapper feature selection process using Support Vector Machine (SVM) classifiers. Results on a dataset of Schizophrenia patients with and without a history of AH are provided which are in good agreement with the models in the literature.