Image Processing Reference
In-Depth Information
image data, and how to interact with the models if, after all, manual editing is
required. Still, hierarchical model representation is an active and challenging field
in 3-D medical image segmentation research in which several investigators have
presented encouraging results in cardiac [7-9,116,122,135,136] and thorax mod-
eling [148,149,160,231].
9.5.2.2
Robustness and Effective Automation
Processing prior to model recovery, automation of the recovery algorithm itself,
and the presence of
ad hoc
parameters are factors that determine the robustness
of a technique and its
effective automation
. By effective automation we refer to
the automation of the overall approach, from raw images until the presentation
of the functional parameters.
Before a given model can be fitted or deformed to a data set, almost every
technique requires some type of preprocessing to convert the raw gray-level
images into a representation suitable for shape recovery. Section 9.4 has suggested
a classification of types of input data. For the sake of simplicity,
Table 9.2
only
indicates the degree of manual involvement required to obtain the corresponding
input data. Four categories were considered: no preprocessing required (N),
manual initialization of landmarks or models (I), (semi) automated initialization
of landmarks and models integrated into the technique (A), and fully manual
segmentation of landmarks and contours (M). Although variability inherent to
the preprocessing can have a marked effect on the overall performance of a
technique, this factor is usually disregarded in the evaluation of algorithms. A
remarkable exception is the evaluation of MR tag-tracking algorithms using
Monte Carlo analysis to assess the influence of erroneous tag localization in the
recovery of tissue deformation [78,83,177,178,240]. Model initialization is also
related to the issue of preprocessing. Although a few techniques make explicit
mention of the procedure required to initialize the model (cf., e.g., [
5
,
6
,
81
,
145
,
159
,
160
]), model initialization in a 3-D environment can be nontrivial or may
require expert guidance.
Another factor undermining the robustness and reliability of a technique is the
presence of
ad hoc
parameters that have to be set by the user. This can be particularly
problematic when such parameters are highly dependent on a given data set. This
is a known problem, for instance, of many physics-based deformable models for
which several weights must be tuned to balance the smoothing constraints to the
external energy terms. However, in the literature, analysis of sensitivity of the result
to the weighting parameters is mostly missing. In
Table 9.1
, we have classified the
different techniques into two categories according to the presence of user-defined
ad hoc
parameters: no parameters or parameters with corresponding analysis of
sensitivity (
) and parameters for which no sensitivity analysis was performed
(+). The fact that several methods do not present
ad hoc
parameters (
−
) should
not be confounded with overall robustness. Even within the approaches with
quantitative evaluation, many papers in the (
−
) category either require substantial
preprocessing [9,27,28,77,177,178,190] or human guidance [113,107,108,117,
−
Search WWH ::
Custom Search