Biomedical Engineering Reference
In-Depth Information
5.6
General Conclusions
We have described image distortions in CT, MRI, ET, and US that can influence
image registration, both within and across modalities. The nature of each effect
varies between modalities, reflecting the image generation mechanisms.
The problems of image distortion in CT are relatively minor when com-
pared with MRI or 3D-US. This is why CT is often considered the spatial gold
standard in multimodality registration and image-guided intervention
studies. The problems of image distortion in MRI are multiple: geometric dis-
tortions due to imperfections in the scanner fields (static, gradients, and RF)
and to interactions between the imaged object and the scanner fields. Scanner
technology is constantly evolving, allowing the acquisition of images of
increasing quality. However, new MR scanners, imaging sequences, and
applications are constantly being developed, often giving rise to worse or
even new artifacts, or to new demands on the accuracy of the images. Inter-
ventional MRI is a case in point; although the artifacts encountered generally
belong to the categories described above, their importance is enhanced by the
greater presence of instruments in the field of view and the need for a high
degree of geometric accuracy. The constant expansion of MRI means that the
issues of image artifacts in general and distortions in particular will remain
an active area of research for many years to come. We also note that the use
of postprocessing methods to correct for spatial inaccuracies in MRI is greater
than in all other modalities. This may reflect both the magnitude of the prob-
lem and the desire to make MRI the superior modality to visualize and quan-
tify structure. In ET, distortions generally result from fundamental physical
or electronic limitations in the image formation process and often from a
compromise between sensitivity and accuracy. In US, the great complexity of
the interactions between the sound waves and the imaged tissue, resulting in
the sometimes dramatic breakdown of the assumptions on which the method
relies, imposes fundamental limitations on the achievable faithfulness of the
reconstructed image. The amplitude of these geometric inaccuracies com-
pared to other undesirable effects or limitations has resulted in a relative lack
of interest in possible correction methods for US.
Using spatial resolution as a criterion, CT and MRI are generally superior
to ET and US. On the other hand, distortions are the most important in MRI
in relative terms, while CT and ET are often considered distortion free. Fur-
thermore, one could argue that the magnitude of the distortions in US
matches its spatial resolution. The implication of these observations is that
the achievable quality of image registration will depend on the modalities
involved and the aim of the registration. For maximum precision, relative dif-
ferences in distortion can be modeled and integrated in the registration pro-
cess. In general, such a model will be used to correct the most distorted
modality. For applications involving an external coordinate system, the need
to transform between image and world coordinates is likely to be a determin-
ing factor and may require distortion modeling for each modality.
