Image Processing Reference
In-Depth Information
outside the vessel boundaries. This limits direct applicability of some visualization
methods, such as integral lines or particle traces.
Segmentation techniques have been developed to directly segment measured 4D
flow data. These methods are based on the assumption that the flow outside the vessel
boundary exhibits incoherent behavior [ 7 , 35 , 40 ]. Van Pelt et al. [ 45 ] presented
an extension of active surfaces to segment flow. Krishnan et al. [ 20 ] introduce a
segmentation technique based on Finite-Time Lyaponov Exponents (FTLE). The
main drawback of these methods is that they will fail in some pathologies due to the
characteristics of the flow, e.g., areas with slow flow.
Most flow visualization techniques require seeding or region selection as initial-
ization. The main reason for the selection is to avoid the clutter that visualizing the
flow in the full domain supposes. The definition of the seeding region is usually done
by probing in the volume domain, often with the help of segmentation.
Van Pelt et al. [ 46 ] presented a semi-automatic technique to probe cross-sections
of anatomical data avoiding full segmentation (see Fig. 25.5 ). If anatomical data is
not available an option for cross-sectional placement is to use the so called temporal
maximum intensity projection (TMIP). For each voxel position of the TMIP scalar
volume, the maximum speed is determined along the time axis of the 4D flow data.
Hence, each voxel with a bright intensity indicates that a flow velocity with a sub-
stantial speed has occurred there at least once during the cardiac cycle. This probing
method has several drawbacks: it assumes tubular structures, so it is only valid for
vessels, and it does not consider the movement of the vessels during the heart cycle.
In later work, Van Pelt et al. [ 47 ] presented a probing technique to allow fast
qualitative inspection, avoiding full segmentation. The user positions a 3D virtual
probe on the viewing plane with common 2D interaction metaphors. An automatic
fitting of the probe is provided for the third dimension, i.e., the viewing direction.
Using the available velocity information of the measured blood flow field, the probe
is aligned to this field. In particular, the automated fitting aligns the orientation of
the long axis of the virtual probe to be tangential to the average local blood flow
orientation. The probe is the basis for further visualizations (see Fig. 25.9 ).
VanPeltetal.[ 46 ] also investigated different local seeding strategies based on the
vessel center (e.g., radial or circular) concluding that fixed template seeding cannot
accommodate flow variations. Krishnan et al. [ 20 ] presented a seeding strategy based
on the segmentation of flow maps [ 41 ]. Flow maps are based on the end position
of the particle after integration or advection. It is expected that this seeding strategy
will adapt to real flow patterns.
In visualization, focus-and-context approaches are commonly used to avoid clut-
ter. Gasteiger et al. [ 11 ] propose the FlowLens which is a user-defined 2Dmagic lens.
This lens combines flow attributes by showing a different attribute and visualization
within and outside the lens. Additionally, they incorporate a 2.5D lens to enable
probing and slicing through the flow. To simplify the interface, they provide scopes
which are task-based. Each scope consists of pairs of focus vs. context attributes,
and propose visualization templates to represent each pair (see Fig. 25.8 ).
Search WWH ::




Custom Search