Functional Connectivity Network Fusion with Dynamic Thresholding for MCI Diagnosis.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Resting-state functional MRI (rs-fMRI) is a key tool for identifying mild cognitive impairment (MCI).
• Previous methods analyzing rs-fMRI connectivity networks in MCI patients often used fixed thresholds, limiting their effectiveness.
• Machine learning approaches have improved MCI diagnosis but often rely on predetermined, uniform network thresholds.

Purpose of the Study:

• To develop a novel element-wise thresholding strategy for rs-fMRI data.
• To dynamically construct multiple functional brain networks using varying thresholds.
• To improve the accuracy of MCI diagnosis by integrating information from these dynamically generated networks.

Main Methods:

• Proposed an element-wise thresholding strategy to dynamically construct functional brain networks from rs-fMRI data.
• Implemented a network fusion scheme to integrate common and complementary information from multiple dynamically generated networks.
• Utilized support vector machine (SVM) with features extracted from the fused network for MCI classification.

Main Results:

• The proposed element-wise thresholding and network fusion framework significantly improved MCI classification performance.
• Dynamic thresholding captured more nuanced network information compared to uniform thresholding methods.
• The integrated features from the fused network led to superior diagnostic accuracy.

Conclusions:

• The novel element-wise thresholding strategy offers a more effective approach to analyzing rs-fMRI data for MCI detection.
• Dynamic network construction and fusion enhance the sensitivity of neuroimaging in identifying cognitive impairment.
• This framework represents a significant advancement in machine learning-based MCI diagnosis using rs-fMRI.