Tensor Based Blind Source Separation in Longitudinal Magnetic Resonance Imaging Analysis.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Data Analysis

Background:

• Investigating white matter (WM) fiber bundles is vital for understanding neurological diseases such as multiple sclerosis (MS).
• The large data volumes in WM analysis necessitate automated processing methods.
• Existing methods may struggle with the complexity of anatomical brain structures.

Purpose of the Study:

• To develop and evaluate an automated, tensor-based approach for modeling and analyzing WM fiber bundles.
• To apply blind source separation (BSS) techniques to formalize WM data using tensor models.
• To differentiate between healthy and pathologically affected fibers within WM bundles.

Main Methods:

• Utilized vector hankelization to create a tensor model from WM fiber-bundle data.
• Applied tensor factorization techniques: (L_r, L_r, 1) block term decomposition (BTD) and canonical polyadic decomposition (CPD).
• Evaluated the model's performance on simulated data representing multiple sclerosis (MS) pathology.

Main Results:

• The tensor-based model successfully formalized complex anatomical brain structures.
• Block term decomposition (BTD) and canonical polyadic decomposition (CPD) effectively extracted information from the tensor model.
• The model demonstrated capability in differentiating between normal and pathologically affected WM fibers.
• Successfully detected subtle pathological phenomena along WM fibers in simulated MS data.

Conclusions:

• Tensor-based blind source separation (BSS) offers a powerful automated approach for analyzing white matter (WM) fiber bundles.
• The proposed method, utilizing tensor factorization, can effectively model complex brain anatomy and identify early signs of neurological disease.
• This technique shows promise for the investigation of white matter pathologies like multiple sclerosis (MS).