Coupling brain-tumor biophysical models and diffeomorphic image registration.

Area of Science:

• Medical Imaging
• Computational Biology
• Biophysics

Background:

• Glioblastoma multiforme presents significant challenges in treatment planning due to complex tumor growth and anatomical changes.
• Accurate analysis of brain tumor MR images requires integrating image registration with biophysical modeling.
• Existing methods often treat image registration and biophysical inversion as separate problems, limiting comprehensive analysis.

Purpose of the Study:

• To present SIBIA (Scalable Integrated Biophysics-based Image Analysis), a framework for joint image registration and biophysical inversion.
• To apply SIBIA to analyze MR images of glioblastomas, addressing normal-to-abnormal image registration and biophysical inversion from patient data.
• To solve the coupled, non-linear, non-convex optimization problem of glioblastoma growth modeling and image registration.

Main Methods:

• Developed a PDE-constrained optimization formulation for the coupled image registration and biophysical inversion problem.
• Implemented an iterative Picard scheme to solve the non-linear, non-convex optimization problem.
• Utilized large-deformation diffeomorphic registration parameterized by Eulerian velocity fields and a reaction-diffusion tumor growth model.

Main Results:

• Demonstrated the convergence of the SIBIA optimization solver on synthetic and clinical datasets.
• Successfully solved the coupled problem in 3D (256^3 resolution) within minutes using 11 compute nodes.
• Showcased accurate tumor growth parameter estimation and registration map generation.

Conclusions:

• SIBIA provides an effective framework for the coupled analysis of image registration and biophysical inversion in glioblastoma.
• The method accurately models tumor growth and registration, offering potential for improved clinical applications.
• SIBIA's computational efficiency enables rapid analysis of complex 3D brain tumor MR images.