Improving parameter estimation in Dynamic Casual Modeling with Artificial Bee Colony optimization

Abstract

Dynamic Causal Modeling (DCM) for fMRI was first proposed to estimate brain connectivity from fMRI data. However, the parameter estimation with Expectation Maximization (EM) method in DCM is prone to local optima. To improve the performance of parameter estimation, this study proposed a hybrid method that integrates the concept of Artificial Bee Colony (ABC) optimization with generic EM used in DCM. From the investigation on real fMRI dataset, the results can indicate that the proposed method could provide higher opportunity to avoid local optimal solution and obtain better final outputs when compared with generic EM. ABC-EM has shown the potential to be a candidate algorithm for DCM estimate brain connectivity for complex experimental tasks involving large number of brain regions and stimuli. Even though the computation time may be concerned, the design of ABC-EM can support parallel computing. The use of ABC-EM on parallel computing system could reduce the computation time.