next up previous contents
Next: k-NN classification and k-distance Up: Application: Camera path-planning in Previous: Path centering

Experimental results

The method proposed in sections [*] and [*] was applied on real medical data, for virtual endoscopies of the aorta and of the colon. The aorta is a relatively smooth structure, but it contains several branches. The colon is a very complex structure, but has a simple tubular topology.

  
Figure 9.5: Computed Tomography of the abdomen. The aorta is shown using a white arrow in slice 25.
\begin{figure}\centerline{\epsfxsize=12cm \epsfbox{figures/chapter4/CT.eps}} \end{figure}

The aortascopy was performed on a structure manually segmented from the abdominal CT of figure [*] by doctors at the Surgical Planning Laboratory - Boston, MA - as part of their abdominal atlas. The CT image, and the atlas derived from it, is made of 114 slices of $256 \times 256$ voxels, which requires to adapt the algorithm of section [*] to 3-dimensional anisotropic data. This does not present any major difficulty.

  
Figure 9.6: Virtual endoscopy of the aorta: camera path generated by an algorithm similar to Lengyel's
\begin{figure}\centerline{\epsfxsize=6cm \epsfbox{figures/endo/aortalengyel.ps} } \end{figure}

The typical result generated by an algorithm such as Lengyel's is illustrated at figure [*]. The path is composed of segments in a reduced set of directions, which makes it irregular. It touches the edges of the structure in all turns in order to keep the path as short as possible.

  
Figure 9.7: Virtual endoscopy of the aorta. Left: Shortest path Right: Path centered and smoothed by the snake energy minimization.
\begin{figure}\centerline{ \epsfxsize=6cm \epsfbox{figures/endo/aortaf1.ps} \epsfxsize=6cm \epsfbox{figures/endo/aortasnake.ps} } \end{figure}

The results of our algorithm are found in figure [*]. The paths show similar qualities to those computed on synthetic 2D data at figure [*]. The path on the left is the shortest possible. It is relatively smooth thanks to the variety of possible segment directions. The path on the right was smoothed and centered.

  
Figure 9.8: Computed Tomography of the colon.
\begin{figure}\centerline{\epsfxsize=12cm \epsfbox{figures/endo/CT2.eps}} \end{figure}

The colonoscopy was performed using the CT scan of figure [*]. Because it was filled with air, the colon appears as a darker structure, which makes it possible to segment using a simple threshold. The segmentation and generation of the 3D model from the CT volume were provided by Dr. Shigeo Okuda, of the SPL.

  
Figure 9.9: Left: Shortest path Right: Centered path flying through the colon
\begin{figure}\centerline{ \epsfxsize=5,5cm \epsfbox{figures/endo/colonfast.ps} \epsfxsize=5,5cm \epsfbox{figures/endo/colonsnake.ps} } \end{figure}

The paths computed by our method are shown at figure [*]. Once again, we show both the shortest and the centered paths. This time, one can see very abrupt changes in directions on the shortest path where the colon turns. Those abrupt changes are smoothed on the centered path.

  
Figure 9.10: Endoscopic view
\begin{figure}\centerline{ \epsfxsize=5,5cm \epsfbox{figures/endo/colonview.ps}} \end{figure}

Finally, a typical colonoscopic view is illustrated at figure [*]. Unfortunately, the paper medium does not allow us to show the dynamic visual effect of the fly-through.
next up previous contents
Next: k-NN classification and k-distance Up: Application: Camera path-planning in Previous: Path centering
Olivier Cuisenaire
1999-10-05