Estimating brain effective connectivity from time series using recurrent neural networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Effective Connectivity (EC) is crucial for understanding brain function and dysfunction in mental illnesses.
• Current EC estimation methods struggle with deep feature extraction from fMRI data.

Purpose of the Study:

• To develop a novel model, Time Series to Effective Connectivity (TS2EC), for predicting Effective Connectivity Networks (ECN) directly from fMRI time series.
• To address the limitations of existing EC estimation methods in deep feature extraction.

Main Methods:

• Proposed a recurrent neural network-based model (TS2EC) for direct EC prediction from fMRI time series.
• Introduced a novel method for generating EC labels using electrocortical stimulation fMRI (es-fMRI) data.
• Evaluated TS2EC on es-fMRI and simulated datasets.

Main Results:

• TS2EC achieved a mean squared error of 0.0057 on the es-fMRI dataset, outperforming existing methods.
• Demonstrated superior performance in accuracy, recall, structural Hamming distance, and F1-score on simulated datasets.
• Significantly higher EC prediction performance compared to current EC analysis methods.

Conclusions:

• TS2EC is a promising novel tool for analyzing Effective Connectivity Networks (ECN) in the brain.
• The model effectively extracts deep features from fMRI data without relying on a fixed model order.
• The use of es-fMRI for EC estimation represents a novel approach in the field.