Manual retrospective methods

Retrospective methods do not require external markers, but use the anatomy itself to search for the best transformation.
The simplest methods rely on the user to manually define the parameters of the registration transformation. For instance, Kapouleas [88] registers MRI and PET scans of the brain by first identifying the inter-hemispheric plane, then interactively adjusting the remaining two translational and one rotational parameters of the rigid transform, while the operator visualizes the edges of the MRI overlaid on the PET data. A similar approach is used by Bohm [8] to iteratively select the parameters of a second degree polynomial transform that registers a computerized brain atlas to PET or MRI.
Alternatively, point-based methods require the user to select corresponding points in both images, and automatically compute the best transformation based on those pairs of corresponding points. Pietrzyk [121] proposes an iterative method based on the repeated selection of one or two such pairs. Hill [76] and Henri [74] select a set of 4 to 26 corresponding points, and compute the transformation that minimizes the global misfit on these.
Manual methods are of course time-consuming and their accuracy is limited to the precision with which the operator can designate matching features.