Anatomical Atlas Generation

Computes an anatomical atlas based on a set of patient images using registration methods. This workflow is automatic and does not require user interaction. The anatomical atlas is composed of a mean image and the set of transformations for each patient's between the mean image and the patient image. The atlas can be used to segment the mean image once and propagate the segmentation results to each patient image automatically.

This workflow has been tested using GIMIAS-1.5 (open source) with a set of 10 patient images and 10 iterations.

Notes:

- Some thresholds are configured as fixed values. Depending on the input image, you should change these.

Technical details: The spatial normalization follows the classic iterative approach presented by Guimond et al. [2]. No intensity normalization is done. For the registration, we use diffeomorphic transformations [4], which are invertible. We cannot blindly average diffeomorphic transformations; the result would not be diffeomorphic [6]. Therefore we use Arsigny’s log-euclidean framework [1], which makes the whole process a lot more complex

[1] Arsigny, Commowick, Pennec, Ayache. “A log-euclidean framework for statistics on diffeomorphisms,” Proc. MICCAI 2006, LNCS vol. 4190, pp. 924-931. Doi: 10.1007/11866565_113

[2] Guimond, Meunier, Thirion. “Average brain models: A convergence study,” Computer Vision and Image Understanding 77(2):192-210, February 2000. Doi: 10.1006/cviu.1999.0815

[4] Rueckert, Aljabar, Heckemann, Hajnal, Hammers. “Diffeomorphic registration using B-splines,” Proc. MICCAI 2006, LNCS vol. 4191, pp. 702-709. Doi: 10.1007/11866763_86

[6] Trouvé. “Diffeomorphisms groups and pattern matching in image analysis,” International Journal of Computer Vision 28(3):213-221, July 1998. Doi: 10.1023/A:1008001603737