Semi-automatic determination of detailed vertebral shape from lumbar radiographs using active appearance models

Abstract

Vertebral body shape annotation can be substantially automated on lumbar radiographs. However, an occasional manual correction may be required, typically with either severe fractures, or when there is a high degree of projectional tilting or scoliosis. The located detailed shapes may enable the development of more powerful quantitative classifiers of osteoporotic vertebral fracture.

Quantitative vertebral morphometry based on parametric modeling of vertebral bodies in 3D

“…The classification outcomes are reported by the true positive rate, true negative rate, area under the curve, and rates of correctly identified morphological cases (wedge, (bi)concavity, and crush) and grades (grade 1, grade 2, and grade 3) of vertebral body fractures TPR true positive rate, TNR true negative rate, AUC area under the curve was aligned to each observed normal or fractured vertebral body in volumetric CT images, its morphometric features were quantitatively described by the shape parameters of the 3D model. In comparison to computerized QM methods based on statistical vertebral body models [15][16][17][18][19][20][21][22][23], where the change of one parameter usually causes several shape deformations, the described 3D parametric model associates each parameter with a specific shape deformation, therefore representing a specific morphometric feature. Another disadvantage of statistical models is that they are generated from examples of vertebral bodies that form a training set and are therefore able to describe only shape deformations that are present in the training set.…”

“…A similar approach was presented by de Bruijne et al [16], who considered shape variations in a population for individual vertebral bodies as well as shape variations among vertebral bodies of the same subject. Brett et al [17] and Roberts et al [18][19][20][21] incorporated shape and texture information into a single statistical model of appearance that was matched to vertebral bodies in radiographic images. On the other hand, Kim et al [22,23] used commercial model-based shape recognition software that semiautomatically determined vertebral body heights, height ratios, and contours from sagittal computed tomography (CT) scout views.…”

The lumbar spine has an intrinsic shape specific to each individual that remains a characteristic throughout flexion and extension

“…Shape modelling of the lumbar spine has been shown to be a reliable, precise [3] and accurate [2] method of characterising sagittal spinal shape. It has also been used to assess vertebral morphometry and fractures in lateral DXA images and vertebral shape on lumbar radiographs [4,5].…”

Active Shape and Appearance Models

“…1 for an example of the appearance of a vertebra in a DXA image). However, by using statistical models of shape and appearance it is possible to accurately and reliably segment such structures [25,31,32].…”