Construction of embedded fMRI resting-state functional connectivity networks using manifold learning

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Resting-state functional magnetic resonance imaging (rsfMRI) is crucial for understanding brain function.
• Schizophrenia is associated with altered functional connectivity networks (FCN).
• Manifold learning algorithms offer advanced methods for analyzing complex neuroimaging data.

Purpose of the Study:

• To construct embedded functional connectivity networks (FCN) using various manifold learning algorithms.
• To evaluate the classification potential of these FCNs for distinguishing schizophrenia patients from healthy controls.
• To compare the performance of different metrics for FCN construction.

Main Methods:

• Applied linear and nonlinear manifold learning algorithms (Multidimensional Scaling, Isometric Feature Mapping, Diffusion Maps, Locally Linear Embedding, kernel PCA) to rsfMRI data.
• Utilized graph-theoretic properties of embedded FCNs for machine learning classification.
• Assessed Euclidean distance and cross-correlation metrics for FCN construction.

Main Results:

• Diffusion Maps combined with the cross-correlation metric demonstrated superior performance in classifying functional connectivity networks.
• This combination outperformed other manifold learning algorithms and metric pairings.
• The graph-theoretic properties of embedded FCNs were effective for distinguishing patient groups.

Conclusions:

• Diffusion Maps with cross-correlation offer a robust method for constructing and analyzing functional connectivity networks from rsfMRI data.
• This approach holds promise for improving the diagnostic capabilities in schizophrenia research.
• The study highlights the importance of selecting appropriate manifold learning algorithms and metrics for neuroimaging analysis.