Vertebrae segmentation in 3D CT data: a review of methods and evaluation approaches

Robust and accurate segmentation of the spine subdivided into individual vertebrae is necessary for subsequent diagnosis of illnesses related to the spine, particularly those requiring detection and classification of bone lesions. Based on correct vertebra segmentation, the current status of a disease under treatment ― as well as its progress ― can be determined and followed. The problem is complicated by frequent heavy deformations of both the spine axis and individual vertebrae due to illness, so that some vertebrae may differ substantially from expected shapes or even be missing. This overview summarises and discusses so far published methods for spine and vertebrae segmentation in 3D CT thoracic data. It suggests a classification of these algorithms based on the used approaches, complexity of algorithms, as well as on achieved efficiencies.