18.07.2014 Views

Segmentation of 3D Tubular Tree Structures in Medical Images ...

Segmentation of 3D Tubular Tree Structures in Medical Images ...

Segmentation of 3D Tubular Tree Structures in Medical Images ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

8.2. Directions for Future Work 129<br />

sors” [65], “tensor vot<strong>in</strong>g” [139], “vessel enhanc<strong>in</strong>g diffusion” [95], or similar techniques.<br />

The last po<strong>in</strong>t is that airways can be characterized locally not only as a dark structure<br />

surrounded by brighter tissue as assumed by our methods <strong>in</strong> Section 7, but they are surrounded<br />

by an airway wall <strong>of</strong> a certa<strong>in</strong> thickness. Incorporat<strong>in</strong>g properties <strong>of</strong> this airway<br />

wall would enhance the selective <strong>of</strong> the airway extraction.<br />

Visualization and user <strong>in</strong>teraction: Although a fully automated segmentation <strong>of</strong> the<br />

tubular tree structures without the need for any user <strong>in</strong>teraction is desirable, there may<br />

still rema<strong>in</strong> cases <strong>in</strong> cl<strong>in</strong>ical practice where fully automated methods may fail. Therefore,<br />

to be cl<strong>in</strong>ically applicable, a system also has to be able to cope with such cases <strong>in</strong> a semiautomated<br />

way. Our developed general approach performs a bottom-up identification<br />

<strong>of</strong> all tube-like structures <strong>in</strong> a given volume <strong>of</strong> <strong>in</strong>terest, such that the overall system is<br />

aware <strong>of</strong> potential problems (e.g. identified but unconnected sub-trees). Comb<strong>in</strong>ed with an<br />

appropriate <strong>3D</strong> visualization this allows giv<strong>in</strong>g a user feedback about potential errors. The<br />

user could then easily verify such situations and <strong>in</strong> case <strong>of</strong> errors manual correction could<br />

be used as such cases occur only very rarely. This behavior <strong>of</strong> our general approach that<br />

allows handl<strong>in</strong>g <strong>of</strong> such errors (unconnected sub-trees) is <strong>in</strong> contrast to other conventional<br />

tubular tree structure segmentation approaches that simply discard the rema<strong>in</strong><strong>in</strong>g tree<br />

parts, which is a severe drawback <strong>of</strong> conventional tube extraction schemes.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!