Between 2009 and 2012, I worked in collaboration with Dr. Anja Schunke and Prof. Diethard Tautz of the Max Planck Institute for Evolutionary Biology, Ploen, Germany. The Institute studies the relationships between genetics, morphology and phylogeny in rodents. Morphological analysis of rodent skulls was performed by painstakingly annotating 3D landmarks on micro-CT image volumes, followed by Procrustes Analysis and Principal Component Analysis to extract the main modes of shape variation. The length of time required to manually annotate tens of points on each image in 3D proved to be the major limitation on how many images could be used in such studies, and thus on the statistical significance of the results.
The Max Planck Society funded a collaborative project with the TINA group at the University of Manchester, entitled "Automatic Identification of 3D Landmarks in Micro-CT Mouse Skull Data", in order to solve this problem. During this project, I added a new GTK+ 2 interface to the software, added 3D volume rendering capabilities, and full color image loading, processing and display. In collaboration with the end-users, I identified functionality that would accelerate manual landmark annotation, and incorporated this into an interface that supported annotation on 3D volume rendered images. This was used to generate training data for a fully automatic annotation algorithm.
Since the aim of the work was to support studies of shape, no shape based constraints, such as statistical shape models, could be used in the automatic annotation algorithm. To do so would risk introducing bias into the results. Therefore, I developed an algorithm based on a coarse-to-fine affine registration of image patches from a database of training images to a query image. Each training image represented a model being fitted to the data. The use of multiple, independent models in this way allowed the development of a very reliable error estimation algorithm, to identify poorly located points. I have continued to develop the concept of using multiple, independent models to sample and thus quantify systematic errors in more recent work.
Manual annotation of 50 landmark points on a volume rendered micro-CT image of a Mus Musculus skull.
Automatic annotation of 50 landmark points on a volume rendered micro-CT image of a Mus Musculus skull. The chequered red-and-green points are those identified as poor fits by the error estimation algorithm.
The result of the project was the TINA Geometric Morphometrics Toolkit. The main aspects of the research are described in three papers published in Frontiers in Zoology:
Bromiley, P.A., Schunke, A.C, Ragheb, H., Thacker, N.A., and Tautz, D.
Semi-automatic Landmark Point Annotation for Geometric Morphometrics.
Frontiers in Zoology 11(61), 2014. [Google Scholar]
Ragheb, H., Thacker, N.A., Bromiley, P.A., Tautz, D., and Schunke, A.C.
Quantitative Shape Analysis with Weighted Covariance Estimates for Increased Statistical Efficiency.
Frontiers in Zoology 10(16), 2013. [Google Scholar]
Schunke, A.C., Bromiley, P.A., Tautz, D. and Thacker, N.A.
TINA Manual Landmarking Tool: Software for the precise digitization of 3D landmarks.
Frontiers in Zoology 9(6), 2012. [Google Scholar]
Whilst the project ended in 2012, the software is still extensively used both within the institute and in other research groups around the world. As with all of the TINA software, it is open-source and can be freely downloaded from the TINA website.