A Survey of Three-Dimensional Modeling Techniques for Quantitative Functional Analysis of Cardiac Images - Advance Image Processing in Magnetic Resonance Imaging - page 305

9.5 DISCUSSION

Comparison of the performance of different techniques is a difficult task due to
the diversity of approaches, the different or complementary information obtained
from them, the varying imaging modalities and image acquisition protocols, and,
last but not least, the lack of a standard way to report performance. In order to
draw some comparative conclusions, we have classified the existing methodolo-
gies according to the degree of their validation (Subsection 9.5.1). At the same
time, we have introduced a number of performance criteria (Subsection 9.5.2).
In this comparison we have focused on techniques leading to traditional cardiac
indices, viz , global (Subsection 9.3.1) and motion parameters (Subsection 9.3.2).

Table 9.2 summarizes this discussion.

9.5.1 V ALIDATION

Three main groups of papers can be distinguished: (1) those with no evaluation
or only qualitative illustrations, (2) those with quantitative evaluation on nonhu-
man data sets, and (3) those with quantitative evaluation on human data sets. This
classification has been used in constructing Table 9.2.
Although there are always exceptions confirming the rule, Table 9.2 indicates
several trends. Most papers in the first category correspond to articles presenting
technical or methodological aspects of advanced modeling techniques. The results
sections in these papers are restricted to either technical aspects or proof-of-
concept illustration on realistic images, indicating the potential of the technique.
Only a few of them have had follow-up articles confirming those hypotheses in
large studies. Further evaluation of these techniques is required in order to deter-
mine their usefulness in clinical tasks.
Approaches in the second category are numerous. Methodologies in this
category have been evaluated on simulated images or in phantom experiments.
These have the advantage of providing the ground truth to assess the accuracy
and reproducibility of the techniques. Owing to the use of idealized geometries
and measurement conditions, extrapolation of the results to in vivo human studies
remains to be demonstrated. Some papers in this second category have evaluated
their techniques on ex vivo or in vivo animal models. Several researchers have
reported experiments with dogs [99,110,115,117,120-122,137,140,142,159,162],
swine [110,117,121,178], or calfs [113,241,242]. * Only a few studies have com-
pared measurements, obtained from ex vivo [241,242] or in vivo [110,142,243,
248] animal studies, with other standard-of-reference techniques.

* Both fitting steps handle sparse data and, therefore, regularization is needed. Regularization, however,
is known to introduce artifactual strains. The effect of three regularization terms has been studied in
[77]: (i) a thin-plate spline stabilizer, (ii) a global smoothing regularizer minimizing the deformation
gradient tensor, F , and (iii) a local body regularizer minimizing the deformation gradient tensor expressed
in some natural local coordinate system (e.g., aligned in circumferential, longitudinal, and radial direc-
tions). Based on simulations of an axis-symmetric deformation of a thick-walled incompressible cylinder,
it was shown that all three constraints yield similar results in the strain analysis.