PIXEL2MECHANICS: AUTOMATED FINITE ELEMENT BIOMECHANICAL SIMULATIONS OF HIGH-RESOLUTION INTERVERTEBRAL DISCS FROM ANISOTROPIC MRIs

IVD degeneration challenges diagnosis and treatment personalization. Biomechanical simulations offer insights into functional mechanobiology but require complex workflows to generate suitable meshes from clinical MRI data. Pixel2Mechanics 1 proposes a novel pipeline for biomechanical FE simulation of IVD meshes directly from low-resolution clinical MRI, aiming to improve personalized IVD modeling 2 . Utilizing a geometrical deep learning framework with cross-level feature fusion and optimization based on differentiable rendering, we generate high-resolution meshes of the lumbar Annulus Fibrosus (AF) and Nucleus Pulposus (NP) from L1-L2 to L4-L5 IVDs. A custom morphing algorithm based on the Bayesian Coherent Point Drift++ 3 generates calibrated volumetric FE meshes 2,4 . Simulations were conducted under daily load conditions 5 to evaluate mechanical responses, comparing these with manual segmentation outcomes. A set of five IVD models with heights ranging from ∼8-16 mm were selected for this study. Pixel2Mechanics outperforms Voxel2Mesh in mesh reconstruction accuracy, with lower average Hausdorff and Point-to-Surface distances (15% and 20% of improvement). Biomechanical simulation comparisons reveal over 88% similarity with manually segmented models (Table 1), showing the same trend: as average height increases, maximum stress increases, while minimum stress decreases. Pixel2Mechanics offers a highly scalable, fully automatic pipeline to generate high-resolution reconstructions of the IVDs from clinical MRI and perform patient-personalized biomechanical simulations for L1-L2 to L4-L5 IVD levels. While triangle meshes effectively represent anatomy, they pose challenges in differentiable rendering. We addressed this with geometry regularizers, but balancing smoothness and rendering remains an open question. Results are highly correlated and agree with those generated by manual segmentations without human-in-the-loop correction. Our tool facilitate the exploration of IVD degeneration while minimizing manual effort, being a valuable tool for future clinical integration, diagnosis, and treatment of personalization for IVD degeneration. FUNDING. EU-Horizon-2020 (MSCA-2020-ITN-ETN-GA:955735), ERC (ERC-2021-CoG-O-Health-101044828), ICREA-Acadèmia.