• Research Home
• Locally Most Similar Atlas Construction
• Robust Detection of DTI Differences
• Design and Use of Atlases for Radiotherapy
• Head and Neck Atlas Construction
• Deformability Statistics in Registration
• Locally Affine Registration
• Registration and Pathologies

Registration in the Presence of Pathological Structures

Warping an anatomical atlas towards a patient image allows the simultaneous segmentation of structures, for example in the context of radiotherapy treatment planning. This is particularly interesting for brain images, since they contain a large number of small but important structures (optical nerves, grey nuclei, etc.). However, the variability induced by the presence of a tumor, a surgical resection or more generally lesions that are not present in the digital atlas, prevents an accurate registration between the atlas and the patient images.

To tackle this problem, we have introduced in [1] a general method to take into account a lesion mask in the registration algorithm. Although this article was developed for a specific registration method, its principle is applicable with minor modifications to any registration method. The mask may be obtained using either manual delineation or automatic methods based on the detection of outlier intensities in the joint histograms of several modalities [1,2]. Taking into account this mask in the registration then amounts to decrease the confidence in the matchings corresponding to the lesion mask and therefore computing the deformation in the lesions region as an interpolation of the neighboring deformation.

We have applied this registration to atlas-based segmentation of brain images for radiotherapy planning in two cases: first when a large tumor is present, allowing to obtain deformations much more realistic (see Fig. 1); and in the case of a surgical resection of a tumor , where our method allows for a much better segmentation of the surrounding regions (see Fig. 2).

Figure 1: Segmentation of a patient image containing a large tumor. Top left: slice of atlas MRI (a) and segmentation (b). Top right: patient image (c), and confidence used for the registration (d). The confidence is 0 inside the tumor (in black on the image). Bottom line: transformation of the atlas segmentation into the patient geometry, by simple registration (e), or by taking into account the tumor (f). Fig. (g) and (h) present zooms on the same area of interest from figures (e) and (f).

Figure 2: Segmentation of a patient image containing a surgical resection. (a) Patient image. (b) Confidence (resection is in black). (c) Result produced by a simple registration, un-aware of the resection. (d) Result produced by our algorithm, exhibiting a better segmentation of the cerebellum (see white arrows).

Bibliography

1. Non-Rigid atlas to subject registration with pathologies for conformal radiotherapy
   R. Stefanescu, O. Commowick, G. Malandain, P.-Y. Bondiau, N. Ayache and X. Pennec.
   MICCAI 2004, volume 3216 of LNCS, pages 704-711, Saint-Malo, September 2004. [bibtex-entry]
    
2. Segmentation automatique des tissus cérébraux en présence de structures pathologiques
   O. Commowick, G. Malandain and P.-Y. Bondiau.
   Congrès Société Francaise de Physique Médicale (SFPM), Montpellier, France, Juin 2004. [bibtex-entry]
    

Designed for Safari, Firefox and MSIE. If you encounter any problem for viewing it, please contact me.