Unsupervised contrastive graph learning for resting-state functional MRI analysis and brain disorder detection.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for understanding brain interactions.
• Current machine/deep learning methods for fMRI disease prediction require extensive labeled data, hindering clinical application.
• Labeling fMRI data is labor-intensive and time-consuming, posing a significant bottleneck.

Purpose of the Study:

• To develop an unsupervised contrastive graph learning (UCGL) framework for fMRI-based brain disease analysis.
• To enable accurate disease identification using unlabeled rs-fMRI data.
• To overcome the limitations of data annotation in current machine learning approaches for neuroimaging.

Main Methods:

• Proposed an unsupervised contrastive graph learning (UCGL) framework utilizing a pretext model for informative fMRI representation generation.
• Implemented a bi-level fMRI augmentation strategy to enhance blood-oxygen-level-dependent (BOLD) signals.
• Employed parallel graph convolutional networks within an unsupervised contrastive learning paradigm for feature extraction.

Main Results:

• The UCGL framework demonstrated superior performance compared to state-of-the-art methods on three rs-fMRI datasets.
• Achieved high accuracy in automated diagnosis across diverse brain diseases, including major depressive disorder, autism spectrum disorder, and Alzheimer's disease.
• Successfully performed cross-site and cross-dataset learning tasks, highlighting the generalizability of the proposed method.

Conclusions:

• The UCGL framework offers a powerful, unsupervised approach for brain disease diagnosis using rs-fMRI data.
• This method significantly reduces the reliance on labeled data, making it more practical for clinical settings.
• UCGL advances the field of neuroimaging analysis by providing an efficient and effective tool for disease identification.