Biomedical Engineering Reference
In-Depth Information
points can be registered to the image surface using one of the surface-matching
algorithms described in Chapter 3, such as the iterative closest point (ICP)
algorithm.
24
Rigid movement can occur within a surface of nearly constant curvature
without any penalty to a cost function that minimizes the distance of phys-
ical points to the surface. It is therefore important to include points on
surfaces which have variable and high curvature to provide a good regis-
tration. Combination of both landmark and surface points has been sug-
gested as a way of overcoming this difficulty. It has been shown that a single
landmark improved multiple surface registration accuracy from 1.5 to 1.0 mm
(mean).
25
12.3.5
Intensity-Based Registration
Algorithms that register two images based solely on the gray level intensity
of corresponding voxels, such as maximization of mutual information, have
become the method of choice for 3D-to-3D image alignment. If an intraopera-
tive image is available, such methods may also be applicable to therapy guid-
ance. The relevance of intensity-based registration to intraoperative images
will be examined in the next section.
12.4
Intraoperative Imaging
The value of using a tracked pointer or similar device is limited in that only a
single point is marked at any one time, and any points must be on the surface of
the patient. However, a wealth of potential information is available from real-
time imaging in the operating room that does not suffer from these limitations.
The imaging modalities and methods of alignment to preoperative images will
be considered in this section.
With intraoperative imaging, one needs to address two technical issues—
calibration and registration. Calibration of the imaging device relates the image
data to 3D space. It is important to consider the spatial integrity of the image data
and the accuracy with which calibration can be achieved. A reference object of
known dimensions is generally used to perform calibration. The device may
be tracked by one of the methods described in Section 12.2.1, or may define
the intraoperative coordinate system itself.
For alignment, data extraction is required if features such as landmark points
or lines are to be identified in the intraoperative image. Direct intensity-based
methods circumvent the need for such segmentation, as alignment is achieved
using the image data directly.
With any method, it is important to consider the speed of the algorithm.
Results must be obtained within a few seconds, or at least a minute or two,
to be useful during a surgical procedure. For the intraoperative imaging
