Biomedical Engineering Reference
In-Depth Information
A first issue in need of extensive investigation to reach such an ambitious goal
is
how to extract knowledge out of visual data
[
29
]. Whereas several solutions have
been provided both from academia and industry to interpret medical images (e.g.
segmentation algorithms), there is still no complete methodology for 3D shape un-
derstanding [
30
]. Thanks to state-of-the-art computational tools, the fully 3D mod-
els permit measurements and the extraction of meaningful features which are much
richer than the ones derived by stacks of images, but the capabilities of such com-
putational tools in the medical domain are not fully exploited yet. Such tools would
help also to fit parametric models from scan data as well as motion capture data and
create personalized models of the body part under observation including kinematics.
A second issue to be tackled is
how to enhance the visual model with such knowl-
edge,
i.e. how to accomplish semantic media annotation. Again, while some solutions
exist for 2D models, a stable mark up of 3D shapes is not available. In fact, a crucial
aspect to take into account is that annotation may be related to either an entire object
or its parts; if the target of the annotation is a part of the object, this part has to be
selected with appropriate—manual or (semi)automatic—tools and associated with
textual data. Such operations are not trivial in the 3D case. Another crucial open
problem is how to couple and maintain consistently the annotations with respect to
the many representations of the geometry, possibly at different levels of resolution.
Related to this, another key issue concerns a
comprehensive and formal way to
document such knowledge
for sharing, search and retrieval purposes. There are a
number of general open problems that should be addressed to realize an ideal cata-
loguing and documentation of the life cycle of 3D objects which apply also to the
medical domain. Among those we can mention coding of the data provenance and
version control, effective metadata structures and interoperability. In parallel, a stan-
dardization issue arises on how to create and manage very large digital repositories,
which should involve many actors.
Another issue to tackle is related to the
semantic visualization of a 3D model
,for
which multi-scale visualization is one of the possible options. The biggest challenge
is how to optimize the rendered images of the 3D data and simplify their visual
complexity while retaining the meaningful details visible. This should be done by
taking into account the characteristics of the specific device which will display the
3D content and the portion of related knowledge to show to the user. Two important
connected issues are involved in this challenge. The first one refers to the so-called
semantic zooming
: while under geometric (standard) zooming the view depends on
the physical properties of what is being viewed, the effect of the semantic zoom is
that of changing the representation of the object (or of some of its parts) for different
spatial scales. When zooming away, instead of seeing a scaled down version of an
object, the user may see a different representation which depends on the meaning to
be imparted. This is clearly much more sophisticated than a Level-Of-Detail repre-
sentation and calls for a semantic description of the shape and its parts. The second
aspect, more technical yet meaningful in terms of outreach, concerns the adaptation
of the content to the end device. In this case, compression and resizing algorithms are
needed, which again should preserve the meaningful features according to the model
