Biomedical Engineering Reference
In-Depth Information
depicting soft-tissue anatomy, and photon emission tomography (PET) and sin-
gle photon emission computed tomography (SPECT) show the functional prop-
erty of the organ under study. Before different types of images are fused, these
images have to be registered. After registration, the skeletal structures and ar-
eas of contrast enhancement seen in CT images can be overlaid on MR images,
and likewise, functional lesions detected with PET or SPECT can be viewed in
the context of anatomy imaged with CT or MR. Image registration can also be
applied to multiple data sets obtained with the same modality at different times
for the purpose of quantitative comparison, which increases the precision of
treatment monitoring with serial images.
Another major biomedical application of image registration is to retinal im-
age matching. Retinal or fundus photographs are standard diagnostic tools in
ophthalmology. In the follow-up of age-related macular degeneration, drusen
deposits need to be tracked and compared (see Sbeh
et al
. [1], Rapantzikos
et al
. [2]). Screening of diabetic retinopathy can involve a follow-up over many
years (see [3, 4]). To determine the progression of glaucoma, a series of optic-
nerve-head topographies are assessed and compared [5]. Serial photographs
of the flow of fluorescein dye are also used to determine areas of ischemia,
hemorrhaging, neovascularization, and occlusions in diseases such as diabetic
retinopathy (see [5]). A noise reduction technique is reported for laser scan-
ning ophthalmoscope using image registration [6]. Multimodality registration
is also performed in retinal imaging. In glaucoma diagnosis, for example, the
optic-nerve-head is assessed from color stereo images and the nerve fiber layer
is assessed from red-free images [7]. In retinal analysis, two types of images,
fluorescein images (angiographic images taken under ultra violet light after in-
jection of fluorescein dye) and green images (taken under natural light with a
green filter), are often used for the diagnosis of the gravity of diabetic retinopa-
thy [3]. Physicians often use more than one image to identify a lesion and assess
its seriousness, or base their diagnosis on detection of various image features in
different modality images. To make this comparison and assessment objective,
it is necessary that all images be registered.
The research of image registration has a relatively short history. Due to di-
verse applications, many registration algorithms have been developed from dif-
ferent perspectives. Brown summarized the research work before 1992, mostly
for 2D-2D registration [8]. Since the most important and fruitful application of
image registration is in medical imaging, several authors reviewed registration
