A unified framework for personalized regions selection and functional relation modeling for early MCI identification

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Cognitive Neuroscience

Background:

• Resting-state functional MRI (rs-fMRI) is crucial for studying brain diseases but faces challenges due to unsupervised data and lack of region-specific labels.
• Existing methods often overlook individual neural activity variations, focusing instead on shared group characteristics.

Purpose of the Study:

• To develop a novel framework for identifying early-stage mild cognitive impairment (eMCI) that accounts for individual variability in brain activity.
• To leverage deep learning and reinforcement learning for personalized diagnostic biomarkers.

Main Methods:

• A deep neural network integrating temporal embedding, adaptive ROI selection via reinforcement learning, and graph-based neural networks for functional relation analysis.
• Subject-specific region selection to capture unique neural activity patterns.

Main Results:

• The proposed framework demonstrates superior performance in identifying eMCI compared to conventional methods.
• Neuroscientific interpretations derived from selected ROIs offer insights into eMCI classification.

Conclusions:

• This individualized approach enhances the accuracy of eMCI detection.
• The framework provides a promising tool for personalized diagnosis in neurological disorders.