A Novel Approach for Shape Segmentation of Vertebrae: Decomposition into Anatomical Regions Using 3D Skeletonization

The shape segmentation of vertebrae refers to the process of dividing 3D geometries of vertebrae into their distinct components. Based on their anatomical structures, the 3D geometry can be separated into the vertebral body, laminae, and various processes. This segmentation is crucial for identifying specific regions of the vertebrae, enabling a precise analysis of their morphology. However, due to the complex shape and varying morphology in different spinal regions, it is challenging to derive such segmentations without manual intervention. We propose a novel, fully automated method for shape segmentation of vertebrae, which is able to segment cervical, thoracic and lumbar vertebrae. The proposed approach extracts the vertebral body by analyzing the vertices distance distribution, and then segments the processes based on 3D skeletonization of the vertebral arch. The experiments show that our method achieves superior segmentation results in comparison with other suitable algorithms across all evaluated metrics such as Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Sensitivity, and Specificity, and anatomical segments, with the most notable performance gap observed in the Spinous Process (DSC of 85% vs. 49%) and Transverse Processes (DSC of 81% vs. 56%).Clinical Relevance-The segmentation of anatomical regions provides a basis for patient-specific analysis and biomechanical modelling of the spine as well as developing of patient-specific implants.