Functional Brain Network Classification With Compact Representation of SICE Matrices.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Sparse inverse covariance estimation (SICE) is used for functional brain connectivity modeling.
• Directly using SICE matrices for Alzheimer's disease discrimination is limited by high dimensionality and few training samples.
• SICE matrices possess lower intrinsic dimensionality due to their Riemannian manifold structure and shared human brain connectivity patterns.

Purpose of the Study:

• To develop a more effective method for discriminating Alzheimer's disease from normal controls using SICE-derived brain connectivity.
• To address the limitations of high dimensionality and data scarcity in SICE matrix analysis.
• To create a compact and interpretable representation of brain networks.

Main Methods:

• Employed manifold-based similarity measures and kernel-based Principal Component Analysis (PCA) to extract principal connectivity components.
• Developed a novel preimage estimation algorithm for anatomical interpretability of connectivity components.
• Validated the method using synthetic data and real resting-state fMRI (rs-fMRI) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Main Results:

• The proposed method significantly improves classification accuracy compared to existing approaches.
• The extracted connectivity components provide a more compact and meaningful representation of brain networks.
• The preimage estimation algorithm enhances the anatomical interpretability of the results.

Conclusions:

• The novel manifold-based approach effectively addresses the challenges of SICE matrix analysis for brain connectivity.
• This method offers a promising tool for Alzheimer's disease diagnosis and understanding brain network alterations.
• The enhanced interpretability facilitates deeper insights into SICE-based brain networks in neuroimaging studies.