Dynamic topology analysis for spatial patterns of multifocal lesions on MRI

Area of Science:

• Medical Imaging Analysis
• Network Science
• Computational Topology

Background:

• Analyzing multifocal lesion spatial patterns in MRI is crucial for disease understanding and precision medicine.
• Current methods using feature engineering and deep learning have limitations in exploring lesion topology.
• Network science offers a way to model inter-lesion topology, but graph construction and community quantification remain challenging.

Purpose of the Study:

• To develop a novel Dynamic Topology Analysis framework using persistent homology to explore multifocal lesion patterns.
• To investigate the predictive value of global geometry and local clusters of multifocal lesions.
• To address challenges in informative graph construction and community structure quantification in network science.

Main Methods:

• Proposed a Dynamic Hierarchical Network to build multi-scale topology from sparse to dense networks.
• Introduced K-simplex Filtration for higher-level topological abstraction and community identification.
• Developed Decomposed Community Persistence algorithm to quantify and track dynamic community evolution.

Main Results:

• The framework successfully constructed multi-scale global and community-level topologies.
• Quantified community structures and integrated them with global geometric invariants for topological pattern analysis.
• Achieved high performance in diagnostic differentiation (ROC AUC 0.875) and prognostic prediction (ROC AUC 0.767) for multiple sclerosis.

Conclusions:

• The Dynamic Topology Analysis framework provides a robust method for quantitatively analyzing multifocal lesion spatial patterns.
• The approach outperforms existing persistent homology, feature engineering, and deep learning methods.
• This framework holds significant potential for advancing precision medicine in multifocal diseases like multiple sclerosis.