Up: Title and abstract Previous: VHD as Source of Test Data

Summary

The VHD provides us with anatomical information about a particular individual with an extremely high spatial and colour resolution. This data is ideal for use as an atlas. We have briefly described a physically based non-rigid registration algorithm for which an atlas segmentation is particularly appropriate. This algorithm follows on from our previous work on non-rigid registration 5,6.

In order to test the accuracy of registration algorithms, both rigid and non-rigid, a 'gold standard' is required. This presents a problem when dealing with real scans, since no such standard exists. Attempts have been made to assess the accuracy of registration using, for example, pins inserted into the skull 3. There are still accuracy problems with such methods, however. By creating synthetic scans from the VHD we can create multiple images where the relative transformations are accurately known. Examples of MRI and ultrasound images are given. This technique may provide useful sets of test data for matching algorithms. The combination of both real and synthetic data should help to answer the problem of validation of registration algorithms.

All the techniques mentioned in this paper will require an accurately segmented version of the VHD. Since many other groups are working on this task it seems inappropriate to pursue segmentation ourselves at present. This is very much work in progress and results are few at present. It seems likely that the VHD will have a considerable role to play in the modelling of deformation and assessment of registration methods.

Acknowledgements

This work is supported by the UK EPSRC grant number GR/J90183.

References

1. Christensen C. A., Rabbitt R. D. et al, Topological Properties of Smooth Anatomic Maps, Proc. IPMI, 1995; Kluwer Acad. Publ.: pp101-112.

2. DL Collins, AC Evans et al, Automatic 3D Segmentation of Neuro-Anatomical Structures, Proc. IPMI, 1995; Kluwer Acad. Publ.: pp139-152.

3. J West, JM Fitzpatrick et al, Comparison and evaluation of retrospective intermodality image registration techniques, Medical Imaging 1996: Image processing, Proc. SPIE 2710, 1996: pp332-347.

4. P Neelin, J Crossman et al, Validation of an MRI/PET landmark registration method using 3D simulated PET images and point simulations, 3D Advanced Image Processing in Medicine, proc. IEEE EMBS 1992: pp73-77.

5. PJ Edwards, J Little, DLG Hill, DJ Hawkes, Medical Image Registration Incorporating Deformations, Proc. 6th British Machine Vision Conference, 1995; 2:691-699.

6. J Little, DLG Hill, DJ Hawkes, Deformations Incorporating Rigid Structures Proc. MMBIA, IEEE Comp. Soc. Press, 1996: pp104-113

Up: Title and abstract Previous: VHD as Source of Test Data