Scope and Limitations

Voxel-intensity-based registration methods are generally robust for within-modality registration and require minimal user intervention. As detailed in the chapter "Across-Modality Registration using Intensity-Based Cost Functions," extensions of these methods are also effective for across-modality registration. Although most attention has been focused on the brain as the target object for registration, voxel-intensity-based methods can also be used for registration of images of other organs such as the breast [24]. These techniques have been shown to be robust in the presence of focal changes [19,20] and can even be used with clinical images that include differences in tumor size and contrast enhancement [3]. Spatial transformation models need not be restricted to those required for intrasubject registration; even moderate order nonlinear warps with hundreds of parameters can be optimized using calculus-based minimization of intensity-based cost functions [22]. Use is not limited to three-dimensional images; two-dimensional photographs can also be registered [2].

These methods are most likely to encounter limitations in cases where the amount of anatomic information contained within the images is low or where the visual scene lacks global features. Registration of a simple square to a displaced version of itself is not easily achieved by an intensity-based cost function because the spatial information is limited to the boundaries. Likewise, attempted registration of a page of print to a displaced copy is likely to fail because all of the spatial information is at such a local scale that the correct displacement is not reflected by the derivatives of the cost function. Failures can also occur when performing simple simulations without the addition of noise because interpolation errors, which do not conform to the parabolic approximation, can predominate the derivatives. Movements that occur during the course of image acquisition will not be corrected by this or any other standard registration method.

0 0

Post a comment