Automatic Recognition of fMRI-Derived Functional Networks Using 3-D Convolutional Neural Networks

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Functional magnetic resonance imaging (fMRI) enables reconstruction of numerous interacting brain networks.
• Current methods lack cross-subject correspondence for these networks, hindering comparative analysis.
• Accurate classification of fMRI-derived networks is crucial for cognitive and clinical neuroscience.

Purpose of the Study:

• To develop and evaluate a deep 3-D convolutional neural network (CNN) framework.
• To enable automatic, effective, and accurate classification and recognition of functional brain networks from fMRI data.
• To address challenges posed by network variability and noise in fMRI analysis.

Main Methods:

• Utilized a deep 3-D CNN architecture designed for spatial pattern recognition.
• Applied the framework to functional brain networks reconstructed using sparse coding of whole-brain fMRI signals.
• Validated the approach using fMRI data from the Human Connectome Project.

Main Results:

• The proposed deep 3-D CNN framework demonstrated effective and robust performance in classifying and recognizing functional brain networks.
• The model exhibited high tolerance for mislabeled training data, indicating robustness.
• Successful classification of large numbers of fMRI-derived functional brain networks was achieved.

Conclusions:

• Deep 3-D CNNs offer a powerful new approach for modeling functional connectomes from fMRI data.
• This deep learning method provides a robust solution for automatic brain network classification.
• The findings advance the potential for large-scale analysis of brain function in neuroscience research.