Sparse Graph Representation Learning Based on Reinforcement Learning for Personalized Mild Cognitive Impairment (MCI)...

Area of Science:

• Neuroimaging and Computational Neuroscience
• Artificial Intelligence in Medical Diagnosis

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals intrinsic brain function patterns via functional connectivity networks (FCNs).
• Analyzing FCNs is crucial for identifying biomarkers associated with mild cognitive impairment (MCI).
• Existing sparse representation methods for MCI diagnosis rely on supervised learning, limiting novel solution discovery.

Purpose of the Study:

• To develop an advanced reinforcement learning (RL) framework for improved MCI diagnosis.
• To overcome limitations of supervised learning in FCN analysis for MCI.
• To enable efficient exploration of the FCN construction task and determine optimal sparsity levels.

Main Methods:

• Proposed an RL-based framework employing a divide-and-conquer strategy for subject-specific FCN construction.
• Decomposed the FCN construction task into manageable sub-problems to facilitate efficient exploration.
• Utilized the learned value function to adapt FCN sparsity based on individual characteristics.

Main Results:

• The proposed framework demonstrated superior performance in MCI diagnosis compared to existing methods.
• Effective exploration of the vast search space was achieved through the divide-and-conquer approach.
• Validation on public cohort datasets confirmed the framework's diagnostic capabilities.

Conclusions:

• The advanced RL framework offers a promising, efficient, and effective approach for MCI diagnosis using rs-fMRI derived FCNs.
• The divide-and-conquer strategy and adaptive sparsity determination enhance the analysis of brain connectivity patterns.
• This method advances the application of AI in neurodegenerative disease diagnosis.