Medical Image Processing - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 25

to the same anatomical point are mapped to each other'. Image registration
is used mainly to detect the changes in same types of images. It is mainly
used in remote-sensing applications such as in multi-spectral classification,
environmental monitoring, change detection, weather forecasting and geo-
graphic information systems (GIS). In medical science, it is used in combin-
ing computer tomography and nuclear magnetic resonance (NMR) data to
obtain complete information about the patient, monitoring tumour growth,
comparing patient's data with anatomical atlases and many others.
Suppose ultrasound is taken at different times, and in order to observe any
structural changes in two images, the mapping procedure is used. Detection
is also required at different stages in tumour/lesion growth, before and
after brain stimulus in functional MRI, rest and stress comparison and so
on. Proper integration of data obtained from two images acquired in the
clinical track of events is required, and this procedure is called registration.
Sometimes, patients undergo many MR and CT at different times to study
any changes in these images or x-ray time series to monitor the growth of
specific bones or single-photon emission computed tomography (SPECT) to
compare ictal and inter-ictal images and so on. After registration, fusion
step is required to display the data. There are many image registration
methods that may be classified in many ways. Maintz and Viergever [12]
suggested a classification method based on nine criteria, which are classi-
fied into the following categories.
Dimensionality : It refers to the number of image dimensions involved, which
may be 2D or 3D. The dimension can be spatial or time series, that is, time is
added to it. Spatial dimension can be 2D-2D, 3D-3D or 2D-3D. Time series
of images is that where time is added to the dimension. It is used in monitor-
ing tumour growth or bone growth where images are taken at long intervals
or post-operative monitoring of healing where images are taken at shorter
intervals or evaluation of drug effects taken at various time intervals.
Nature of registration methods : Registration can be extrinsic where an artificial
object is attached to the patient's body. The marker object used may be inva-
sive or non-invasive, but non-invasive markers are less accurate. The follow-
ing are examples of some markers: stereotactic frame that is screwed rigidly
to the patient's outer skull used in neurosurgery [6], screw-mounted marker
[3] and marker glued to the skin [11]. Registration can be intrinsic where the
image information is generated by the patient. It can be as follows:
1. Landmark based where a limited set of identified points are identified
and the accuracy depends on the accurate indication of correspond-
ing landmarks in all modalities using some similarity measures.
2. Segmentation based where segmented structures (commonly object
surfaces) are aligned. Segmentation based can be rigid where the
same structures are extracted from both the images, and these are
used as the input to the alignment process. It can be deformable