A brain effective connection learning method based on causal self-encoder and meta-learning

Abstract

The application discloses a brain effect connection learning method based on a causal autoencoder and meta learning, first encodes input fMRI time sequence data through a causal autoencoder (CAE), extracts time sequence space information, then learns causal connections between brain regions by using a structural equation modeling (SEM) in the CAE, finally, uses a model-agnostic meta-learning (MAML) reptile to learn brain effect connection meta knowledge (refers to brain effect connections commonly possessed by most subjects) shared between different subjects, and migrates to the CAE to improve the recognition ability of the model on small sample data. The application uses meta learning to utilize the brain effect connection information commonly shared between different fMRI subjects, effectively alleviates the problem that small sample data is difficult to provide sufficient data. Experimental results on real resting state fMRI data show that the application can provide an alternative auxiliary means for medical researchers to analyze the mechanism of the human brain.

Images