Thoracic vertebral shape: comparison of scoliotic and non-scoliotic vertebrae using MR images and statistical shape modelling

Aims of the study

We aim to assess shape differences between scoliotic and normal vertebrae. This builds on our previous BSRF funded work to develop reliable methods for comprehensively characterising vertebral shape. We hypothesize that we will identify novel differences that we can subsequently use to understand initiation and progression of scoliosis.


The proposed research builds on our prize winning work [1] to develop methods for characterising vertebral shape. Our methods use statistical shape modelling to characterise shape from magnetic resonance (MR) images. Unlike conventional measurements, which consider one aspect of shape at a time, shape modelling comprehensively captures the shape of whole objects and is ideal for analysing complicated, irregular shapes such as vertebrae.


Our previous work developed and tested the methods on a sample of twenty volunteers; showing them to be reliable and to correctly characterise expected variation in normal vertebral shape [2]. In the proposed research we will apply our methods to a sample of patients with scoliosis so that we can identify differences between normal and scoliotic vertebrae.


Twenty patients with scoliosis (thoracic curve, un-operated) will be recruited from the Royal Devon and Exeter NHS Trust and scanned to acquire MR images of their thoracic vertebrae. We will use our previously developed protocol where each vertebrae is scanned individually, using a T1-weighted sequence, as a stack of slices orientated parallel to the mid-transverse plane of the vertebral body. The scans, acquired using the University of Exeter’s 3T MR scanner, will produce image data with voxels of ~0.5x0.5x2 mm.


The shape model will be created by placing 77 landmark points to define the shape of key vertebral features (vertebral body and canal, pedicles, transverse and spinous processes, inferior and superior facets). Landmark points will be placed for all 12 thoracic vertebra from every patient (i.e. 240 sets of landmark points) and then combined with the points from our previous study on volunteers.


The landmark points will be processed using Procrustes analysis to align and scale the points into a common frame of reference (this removes differences in vertebral size and orientation). The shape model will then be created using principle component analysis to determine the mean shape and the modes of variation that describe patterns in the way the landmark points vary about the mean shape. Each vertebrae will have a score that describes its shape in terms of these modes of variation.

We will then assess the differences between scoliotic and non-scoliotic vertebrae using repeated measures analysis of variance with vertebral level as a within-subject factor. The vertebrae will be treated separately (since vertebral shape differs considerably along the thoracic spine) and also in combination (to determine if there is a systematic difference between people).

The research will identify differences in shape between healthy and scoliotic vertebrae. Because our methods use MR image data they are safe for use on multiple occasions. This means that after completing the proposed research we will have the methods and information required to plan and perform longitudinal studies to investigate relationships between vertebral shape and the initiation and progression of scoliosis. This will lead to a better understanding of the pathogenesis of scoliosis and may also lead to more sensitive and safer monitoring methods; thus addressing several research priorities from the Scoliosis Priority Partnership [3].

Share this page