Learning mechanical inhomogeneity of keloid scars in vivo with shear wave elastography based on physics-informed neural networks

Keloids are characterized by disorganized fibro proliferation that extends beyond the original wound margins, resulting in excessive extracellular matrix (ECM) production. Mechanical imaging of keloids in vivo is crucial for both diagnosis and management but remains challenging due to their irregular geometries and heterogeneous properties. To address these issues, we propose to use shear wave elastography based on physics-informed neural networks (PINNs) to map the spatially-dependent elastic properties of keloids. A key challenge is the free-boundary condition with irregular geometric shape created by keloids. To overcome this challenge, we designed an ultrasound-compatible hydrogel that facilitates in vivo measurements by covering the tested keloid, which transforms the free boundaries into manageable interfaces. Both finite element simulations and phantom experiments were performed to validate the method. Additionally, in vivo experiments confirmed the clinical utility of this imaging approach. Our results demonstrate that the spatial shear modulus of keloid can be successfully inferred using this experimental design and PINN-based elastography. This method enables the creation of elasticity maps for diverse keloid scars, which could improve the diagnosis, treatment, and prevention of keloidogenesis by elucidating the relationship between keloid growth and its elastic properties. Additionally, this approach offers insights into the mechanisms underlying mechanotherapeutic interventions. STATEMENT OF SIGNIFICANCE: Keloid scars are inherent inhomogeneous soft tissue fibrosis with irregular geometries and limited dimensions. Imaging the mechanical inhomogeneity of these soft tissues in vivo finds broad clinical applications but remains a great challenge to date. We successfully address the challenge of imaging the spatially-dependent elastic properties of keloid scars using physics-informed neural networks and an ultrasound hydrogel to manage irregular boundaries. Both finite simulations and phantom experiments have been performed to validate the proposed method. Primary in vivo experiments demonstrate that our method enables clinical assessment of mechanical inhomogeneity, offering potential for diagnosis, treatment, and prevention of keloid progression.