Using statistical shape models to develop generalised models of vertebrae

Within OncoEng, the team at UCL aims to develop state-of-the-art imaging and computational tools to predict the growth of metastatic lesions and how these can result in fractures in the spine. This involves a variety of teams with different goals. All of them require a computational mesh that will allow a number of models to run in tandem.

Therefore, a strategy is being developed to simultaneously have a patient-specific mesh that does not introduce biases to the mathematical models and also ensures that the numerical implementation of the models converge. One promising approach for this purpose is called statistical shape modelling. It consists of collecting a large number of vertebrae meshes that capture the variability of the shapes of the bones in the spine. From this database, a reference mesh is chosen, and then a Gaussian model is fitted so that a close approximation of all the bones in the population can be generated by transforming the reference mesh.

Fig. 1:  An example of a statistical shape model of a vertebra from the work of Clogenson et al. where different philologically correct vertebrae shapes can be generated from the mean by moving the coloured points along the surface, in this example, by 2 standard deviations.

This methodology has been used to study bone shapes in many contexts, and it has already been shown by Clogenson et al. (2014) to allow the creation of a generalised model of cervical vertebrae that captures their morphological complexity across a population. In addition, Pereañez et al. (2015) applied this strategy to lumbar vertebrae and were able to statistically discriminate healthy vertebras from vertebras from patients that present scoliosis and fractures. At OncoEng, we hope to build upon this research and understand how a similar strategy can be used to create a generalised model of vertebrae that presents metastatic tumours.

Blog post written by Simao Laranjeira


Clogenson, Marine, et al. “A statistical shape model of the human second cervical vertebra.” International journal of computer assisted radiology and surgery 10 (2015): 1097-1107.

Pereañez, Marco, et al. “Accurate segmentation of vertebral bodies and processes using statistical shape decomposition and conditional models.” IEEE transactions on medical imaging 34.8 (2015): 1627-1639.