Abstract:Inferring effective connectivity between different brain regions from functional magnetic resonance imaging (fMRI) data is an important advanced study in neuroinformatics in recent years. However, current methods have limited usage in effective connectivity studies due to the high noise and small sample size of fMRI data. In this paper, we propose a novel framework for inferring effective connectivity based on generative adversarial networks (GAN), named as EC-GAN. The proposed framework EC-GAN infers effective connectivity via an adversarial process, in which we simultaneously train two models: a generator and a discriminator. The generator consists of a set of effective connectivity generators based on structural equation models which can generate the fMRI time series of each brain region via effective connectivity. Meanwhile, the discriminator is employed to distinguish between the joint distributions of the real and generated fMRI time series. Experimental results on simulated data show that EC-GAN can better infer effective connectivity compared to other state-of-the-art methods. The real-world experiments indicate that EC-GAN can provide a new and reliable perspective analyzing the effective connectivity of fMRI data.

EC-GAN: Inferring Brain Effective Connectivity via Generative Adversarial Networks