![](bg5.png)
corresponds to a different color in all obtained res-
olutions. According to these results, we notice that
segmentation at low-resolution (10% of the original
data) leads to an extraction of the major surface fea-
tures/characteristics, whereas when the algorithm is
applied to the surface at high resolution (100%), it
does not necessary produces more relevant structure
information but includes small features, which do not
contribute very much to the overall shape. In addi-
tion, low-resolution description can be used to deter-
mine constraints for the segmentation at higher reso-
lution. For instance, these first results are encourag-
ing but further investigation is required to extend the
algorithm to a large range of data.
Figure 6: Segmentation of the cortical surface according to
table 1.
5 CONCLUSIONS
In this paper, an automated approach for brain MRI
segmentation and discrete mesh characterization is
proposed. We have currently explored and outlined
the importance of multiresolution representation to
simplify processing, segment meshes and accelerate
medical analysis. Our technique has a great interest
in the study of structural and functional characteris-
tics of the brain. It is also relatively computationally
efficient. At this stage, we have only applied the ap-
proach to a few experimental cases and we have pre-
sented some preliminary results to demonstrate its po-
tential: the method gives satisfying results for mesh
labelling in the case of multiresolution representation.
Even if they have not yet been compared to manual
or other automatic segmentation results, we think that
they are encouraging and faster than manual proce-
dures. However, there are some future works to do.
Clinical validation remains to be done, which will re-
quire additional work. Future validations will com-
pare our segmentation with manually labelled data
and other segmentation results. Finally, the same
framework can be used and extended to segment and
quantify abnormal brains.
REFERENCES
Dempster, A., Laird, N. M., and Rubin, D. (1977). Maxi-
mum likelihood from in complete data via the em al-
gorithm. Journal of the Royal Statistical Society, 1:1–
38.
Hamman, B. (1993). Curvature approximation for tri-
angulated surfaces. In: Gerald Farin et al., edi-
tors, Geometric Modelling, Computing Supplemen-
tum, Springer-Verlag, Vienna, Austria, pages 139–
153.
Hoppe, H. (1996). Progressive meshes. SIGGRAPH ’96
Proc, pages 99–108.
Jaume, S., Macq, B., and Warfield, S. (2002). Labeling
the brain surface using a deformable multiresolution
mesh. MICCAI, pages 451–458.
Lorensen, W. and Cline, H. (1987). Marching cubes:
A high-resolution 3-d surface construction algorithm.
ACM Comput. Graph, 21:163–170.
Mangan, A. and Whitaker, R. (1999). Partitioning 3d sur-
face meshes using watershed segmentation. IEEE
Transactions on Visualization and Computer Graph-
ics, 5.
Max, N. (1999). Weights for computing vertex normals
from facet normals. Journal of Graphics Tools,
4(2):1–6.
Meyer, M., Desbrun, M., and Schroder, P. (2002). Dis-
crete differential-geometry operators for triangulated
2-manifolds. International Workshop on Visualization
and Mathematics, Berlin, Germany,.
Ronfard, R. and Rossignac, J. (1996). Full-range approxi-
mation of triangulated polyhedra. Computer Graphics
Forum, Proc. Eurographics ’96, 15:1–38.
Rossignac, J. and Borrel, P. (1993). Multi-resolution 3d ap-
proximations for rendering complex scenes. In Falci-
dieno, B. and Kunii, T., editors, Theoretical Aspects of
Computer Software, Modeling in Computer Graphics:
Methods and Applications, pages 455–465.
Schroeder, W., Zarge, J., and Lorensen, W. (July 1992).
Decimation of triangle meshes. Computer Graphics
(SIGGRAPH ’92 Proc.), 26:65–70.
Whitaker, R. (2000). Reducing aliasing artifacts in iso-
surfaces of binary volumes. IEEE Volume Visualiza-
tion and Graphics Symposium, 21:23–32.
MULTIRESOLUTION MESH SEGMENTATION OF MRI BRAIN USING CLASSIFICATION AND DISCRETE
CURVATURE
425