GATE: Graph CCA for Temporal Self-Supervised Learning for Label-Efficient fMRI Analysis.

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Functional magnetic resonance imaging (fMRI) is crucial for neuro-disease classification.
• Graph convolutional neural networks (GCNs) show promise but require extensive labeled data and are sensitive to noise.
• Label-efficient learning is essential for practical fMRI-based disease analysis.

Purpose of the Study:

• To develop a novel self-supervised learning (SSL) framework for GCNs to improve fMRI representation learning and classification.
• To address the challenge of limited labeled data in neuro-imaging studies.
• To enhance the robustness and accuracy of neuro-disease classification using fMRI.

Main Methods:

• Propose Graph CCA for Temporal sElf-supervised learning on fMRI analysis (GATE), a theory-driven SSL framework for GCNs.
• Investigate novel graph augmentation strategies using dynamic functional connectives (FC) from fMRI data for SSL.
• Utilize canonical-correlation analysis (CCA) on temporal embeddings within a two-step GCN learning procedure (SSL followed by fine-tuning).

Main Results:

• The proposed GATE framework demonstrates superior performance in neuro-disease classification tasks.
• Effective feature extraction and classification were achieved in a label-efficient setting.
• The method showed significant improvements in diagnosing autism and dementia on two independent fMRI datasets.

Conclusions:

• GATE offers a robust and effective approach for neuro-disease classification using fMRI, particularly in label-limited scenarios.
• The developed SSL strategy enhances GCNs' ability to learn meaningful representations from fMRI data.
• This framework holds significant potential for advancing diagnostic tools in neurology.