Generation of Patient-Specific, Ligamentoskeletal, Finite Element Meshes for Scoliosis Correction Planning

Finite element (FE) biomechanical studies for adolescent idiopathic scoliosis (AIS) treatments will greatly benefit from utilizing true-scale, patient-specific anatomy that accurately characterizes all tissue properties. This study presents a method to automatically generate patient-specific, FE meshes containing volumetric soft tissues, such as ligaments and cartilage, that are inconspicuous in computed tomography (CT) imaging of AIS patients. Convolutional neural network (CNN) derived vertebrae segmentations, obtained from CT scans, provide a foundation for subsequent elastic deformations of ligamentoskeletal, computer-aided designed (CAD) surface meshes, to ascertain patient-specific anatomy, including soft tissue structures. Patient-specific, ligamentoskeletal meshes are then tetrahedralized for use in FE studies. Dice similarity coefficients of 90% and submillimeter Hausdorff distances demonstrate vertebrae and intervertebral disc fitting accuracy of the automatic methodology. In severe AIS cases, when CNN segmentations fail due to overfitting, a semi-automatic step augments the automatic method. The generated FE meshes can provide the basis for biomechanical simulations seeking to correct AIS through bracing, minimally invasive operations, or patient-specific, surgical procedures, like posterior spinal fusion.