Bones
     For the reconstruction of bones, CT images and anatomical images have been used. Reconstruction of the bones on anatomical cross-sections is easy when they are surrounded by musculature, the white color of the compact bone contrasts well with the red one of the skeletal muscular tissues. In this case, reconstruction with Label, the first method presented in this paper, is almost automatic. Also the method that uses Shape Constrained Deformable Model gives good results and is faster than Label.

     In the case where tendons and fibrous articular capsules cross or attach to bones, delineation of the osseous surface may become difficult. With Label, the user has to define the arbitrary boundary on each slice. Segmentation with Label becomes very slow. The second tool can interpolate missing boundaries as described early. In some case, the interpolation is not precise enough; the user can modify the surface by moving a set of vertices to the contour defined arbitrary. Generally speaking, the second tool is faster than Label for the reconstruction of bones on anatomical images. The first image below shows the segmentation of the humerus with Label. For the next two images we can see the result of reconstruction from anatomical female images with shape constrained deformable model. The generic model used for this reconstruction has been created with Label from anatomical male data (Figure 5, Figure 6).
 
     On CT images, distinction between calcified tissue and non-calcified fibrous structures is easier. Comments on reconstruction from these data are the same than comments on reconstruction from anatomical data. The only difficulty we have encountered is that on these images, bones appear with two contours. These contours are very close to each other. During the reconstruction process with the second tool, some vertices of the deformable model move to the internal contour and others to the external contour. To solve this problem, we need to define an initial model enough bigger to contain the entire organ in the voxmap. In that way, vertices only detect the external contour. The figure below shows an example of a 3D reconstruction of a muscle. The first image represents the 3D model before the fitting and the second image shows the result of the fitting process (Figure 7).

Muscles
     Segmentation of muscles is more difficult than bones. In most cases, muscles are surrounded by other muscles. The distinction between individual muscles is often impossible. Thus, it would be illusory to believe that the contour of every muscle can be clearly and easily identified.

     Reconstruction of muscles with Label is very slow. The user has to use the different fiber orientation on anatomical images. In the case where delineation is impossible, the user has to arbitrarily trace a border between muscles where it is expected based on anatomical experience. The user then reconstructs the supposed muscles and judges the correctness of the segmentation with the aid of 3D rendering. When not satisfied, he/she returns to the slices and improves the segmentation until 3D rendering corresponds to the usual muscle size and shape. This segmentation process is very slow.

     Reconstruction with shape constrained deformable model is easier. As described early, this method uses a generic model to interpolate missing boundaries. The reconstruction gives good results if the initial shape is closed to the shape of the muscle be reconstructed. In that case, extrapolation works fine and the segmentation can be done in a few minutes. The two following images illustrate a reconstruction of the Brachialis with the second tool (Figure 8).
 
Skin
     The segmentation of the skin surface does not impose problems due to the fact that the embedding medium (blue) is in clear contrast to the white color of the skin. With the automatic contouring feature of Label, the reconstruction of skin on anatomical images is very fast. The second tool is not adapted to the segmentation of the skin. The shape of the skin surface depends on the posture of the body. It is impossible to define a generic model of the skin surface. The surface skin can be easily segmented with a technique such as Marching Cubes or Region Growing.