Amortization Transformer for Brain Effective Connectivity Estimation from fMRI Data.

Area of Science:

• Neuroinformatics
• Bioinformatics
• Computational Neuroscience

Background:

• Estimating brain effective connectivity from fMRI data is crucial in neuroinformatics and bioinformatics.
• Current machine learning methods often necessitate subject-specific model retraining, failing to utilize cross-subject knowledge.
• This limitation hinders the efficiency and generalizability of brain network analysis.

Purpose of the Study:

• To propose a novel framework, AT-EC, for estimating brain effective connectivity.
• To develop an amortized model that leverages shared information across multiple subjects.
• To enhance the accuracy of effective connectivity estimation using an assisted learning mechanism.

Main Methods:

• An amortization transformer is employed to model fMRI time series dynamics and infer cross-subject effective connectivity.
• An amortized model is trained to capture shared knowledge from diverse subjects.
• An assisted learning mechanism, utilizing functional connectivity, is integrated to refine effective connectivity estimation.

Main Results:

• The AT-EC framework successfully models fMRI time series dynamics and infers brain effective connectivity.
• The amortized approach effectively leverages shared knowledge across subjects.
• Experimental results on simulated and real-world fMRI data validate the proposed method's efficacy.

Conclusions:

• AT-EC offers a novel and effective approach for estimating brain effective connectivity from fMRI data.
• The framework's ability to learn from multiple subjects improves upon existing methods.
• AT-EC demonstrates significant potential for advancing neuroinformatics and bioinformatics research in brain network analysis.