PhyMorphS: A Physics-Guided Framework for Mesh-Based Morphological Synthesis of White Matter Microstructures

Abstract

Numerical phantoms provide controlled 3D representations of brain tissue microstructure for developing and evaluating biophysical models of diffusion-weighted Magnetic Resonance Imaging (dMRI). Existing phantom synthesizers employ different geometric primitives and optimization strategies to generate densely packed substrates with diverse axonal morphologies. However, direct deformation of axonal surfaces under mechanical constraints remains less explored. We propose PhyMorphS, a mesh-based framework in which axonal morphology emerges through physics-guided growth and interactions with the local microstructural environment, without constraining cross-sections to predefined symmetric shapes. In PhyMorphS, axons, myelin sheaths, and cell bodies are represented by triangle meshes, while a mass-spring system and collision-resolution scheme guide iterative axonal growth under mechanical and packing constraints. This enables irregular, asymmetric, and non-convex cross-sections, longitudinal diameter variation, and user-defined morphological alterations such as axonal beading and myelin layer decompaction. We evaluate PhyMorphS against large-scale X-ray nano-holotomography segmentations of real 3D axons from four white matter regions with distinct fiber organizations. The synthesized axons reproduce key region-specific morphological characteristics and achieve high packing densities, while the parallelized implementation scales approximately linearly with axon count in the generation voxel. By reproducing the geometric variability observed in real WM tissue, PhyMorphS provides controlled substrates for investigating how realistic morphology influences simulated dMRI signals, testing the geometric assumptions of biophysical microstructure models, and generating diverse paired microstructure-signal datasets for training machine-learning models.