The objective of this study was to assess the reliability and appropriateness of statistical shape modelling for capturing variation in thoracic vertebral anatomy for future use in assessing scoliotic vertebral morphology.

Magnetic resonance (MR) images of the thoracic vertebrae were acquired from 20 healthy adults (12 female, 8 male) using a 1.5 T MR scanner (Intera, Philips). A T1 weighted spin-echo sequence (repetition time = 294 ms, echo time = 8 ms, number of signal averages = 3) was used. A set of slices (number = 27, thickness = 1.9 mm, gap = 1.63 mm, pixel size = 0.5 mm) were acquired for each vertebrae, parallel to the mid-transverse plane of the vertebral body. Repeated imaging, including participant repositioning, was performed for T4, T8 and T12 to assess reliability. Landmark points were placed on the images to define anatomical features consisting of the vertebral body and foramen, pedicles, transverse and spinous processes, inferior and superior facets. A statistical shape model was created using software tools developed in MATLAB (R2013a, The MathWorks Inc.). The model was used to determine the mean vertebral shape and ‘modes of variation’ describing patterns in vertebral shape. Analysis of variance was used to test for differences between vertebral levels and subjects and reliability was assessed by determining the within-subject standard deviation from the repeated measurements.

The first three modes of variation, shown below (green = mean, red and blue = ±2 standard deviations about the mean), accounted for 70% of the variation in thoracic vertebral shape (Mode 1 = 44%, Mode 2 = 19%, Mode 3 = 4%). Visual inspection indicated that these modes described variation in anatomical features such as the aspect ratio of the vertebral bodies, width and orientation of the pedicles, and position and orientation of the processes and facet points. Variation in shape along the thoracic spine, characterised by these modes of variation, was consistent with that reported in the literature. Significant differences (p< 0.05) between vertebral levels and between some subjects were found. The reliability of the method was good with low relative error (Mode 1 = 5%, Mode 2 = 8%, Mode 3 = 19%).

Statistical shape modelling provides a reliable method for characterizing many anatomical features of the thoracic vertebrae in a compact number of variables. This is useful for robustly assessing morphological differences between scoliotic and non-scoliotic vertebrae and in assessing entry points and trajectories for pedicle screws.