Automatically recognize and segment morphological features of the 3D vertebra based on topological data analysis

Abstract

Background:For spinal surgery, exact knowledge about the shape of individual vertebra is of great importance. However, due to the complex morphological features of human vertebrae and spine, it is challenging to locate, segment automatically, and recognize the morphological features in vertebral images. Significantly, pedicle recognition is more challenging because of the particular structure. Methods:Topological structures such as the Reeb graph could facilitate effective visualization and interactive exploration of feature-rich data. In this paper, we conducted topological data analysis on the 3D vertebra, whereby some principal morphological features of the 3D vertebra are recognized and segmented. First, a scalar field of the 3D vertebra is created in a vertebra coordinate system (VCS). Then, the Reeb graph is adopted for topological data analysis on the scalar field. Morphological features of the 3D vertebra are separated using a cycle-detect-based algorithm in the Reeb graph, and the valid pedicle region is finally generated. Pedicle morphometry is measured for surgical references. Results:Experiments on the dataset from the CSI 2014 Workshop with our method show that the spinous process and vertebral body are 100% (255/255) recognized, the pedicle is 99.8% (509/510) recognized, the transverse process is 94.1% (240/255) recognized. The parameters incl. chord length and diameter of pedicle morphometry are measured and verify the efficiency of the valid pedicle region deduced from the recognized pedicle. Conclusion:Topological data analysis is an effective and promising automatic tool for segmenting and recognizing morphological features on the 3D vertebra. The final extracted valid pedicle region and its pedicle morphometry can provide good references for pedicle screw placement.