Constrained Tensor Decomposition for Longitudinal Analysis of Diffusion Imaging Data.

Area of Science:

• Medical Image Analysis
• Neuroimaging
• Computational Neuroscience

Background:

• Analyzing complex medical imaging data, particularly diffusion imaging (DI), presents challenges due to heterogeneous information.
• Longitudinal DI data offers valuable insights into the temporal evolution of brain microstructure, crucial for understanding progressive diseases like multiple sclerosis (MS).

Purpose of the Study:

• To develop and validate a tensor-based framework for detecting longitudinal changes in DI data from MS patients.
• To identify specific white matter (WM) fiber bundles and their sections exhibiting pathological changes over time.

Main Methods:

• A three-part framework involving preprocessing of longitudinal DI acquisitions, WM fiber-bundle extraction, data tensorization, and rank selection.
• Application of a parallelized constrained tensor factorization algorithm to identify pathological longitudinal changes.
• Validation using simulated longitudinal variations and real MS patient data.

Main Results:

• The proposed framework demonstrated high accuracy and precision in detecting subtle longitudinal changes.
• The method successfully identified specific fiber subsets and sections within WM bundles affected by pathological changes.
• Effective application on both simulated and real-world MS longitudinal data.

Conclusions:

• The developed tensor-based framework is effective for analyzing longitudinal DI data in MS.
• This approach enables precise detection of pathological changes in WM microstructural integrity over time.
• The findings contribute to a better understanding of MS progression and could aid in treatment monitoring.