MetaCAE: Causal autoencoder with meta-knowledge transfer for brain effective connectivity estimation

Junzhong Ji, Zuozhen Zhang, Lu Han, Jinduo Liu

Abstract: Highlights•To the best of our knowledge, this is the first study to incorporate meta-learning strategies into a causal autoencoder for estimating brain effective connectivity (EC) from small-sample functional magnetic resonance imaging (fMRI) data.•The proposed method is a novel end-to-end causal autoencoder framework that utilizes a temporal convolutional encoder to extract non-stationary temporal features of brain regions and a structural equation model decoder to estimate brain EC from fMRI data.•We propose leveraging a meta-learning bi-loop strategy to learn shared brain EC meta-knowledge across subjects, which can help alleviate the issue of insufficient fMRI data and further improve the accuracy of brain EC estimation.•Systematic experiments conducted on both simulated and real fMRI datasets demonstrate that the proposed method outperforms several state-of-the-art approaches on small-sample fMRI data.•The method presented in this paper is expected to advance the development of brain effective connectivity estimation methods and provide technical support for the better understanding of the intrinsic causal relationships between brain regions.