Graphs of the RMS registration error in Chapter 3 clearly shows that the accuracy of a registration is greatly increased by the use of a large number of landmarks. In the OR context however, the length of time available to perform the registration procedure is limited. Moreover, few natural landmarks are subject to good localization in both spaces. This makes it impractical to use a large number of landmarks for a registration if each of these points must be identified on both the patient and the image.
One approach that could partially solve this problem is the usage of some ``localizer'' that would consist of a large set of small MR-visible markers fixed permanently to a rigid support, and whose relative positions would be known precisely. We could imagine for instance, the localizer as being a sort of ``helmet'' in which the markers are embedded. This helmet could be rigidly mounted on a light-weight ring that would be attached to the patient's head. The separation between the localizer and the attachment (i.e., the ring) would allow it to be removed after the scan is performed.
A careful design of the markers would allow the precise determination of their centres in image space. For instance, Sumanaweera et al. [60] report a technique allowing the determination of the marker localization to a precision better that one tenth of a pixel. Analysis of the field distortion caused by the markers and the support would also increase the localization precision by allowing correction for susceptibility shift. Moreover, marker identification can be performed automatically since the computer knows the marker configuration a priori.
If landmarks, which can be unambiguously reached with the probe and whose positions relative to the set of markers on the helmet are accurately known, are built on the attachment, the registration procedure would be limited to identifying these landmarks on the ring. The known position of the markers on images could be used to correct for the non-linear distortion that can be present in the imaging data. Note that the helmet would be only needed during the image acquisition. Only the attachment would be required for registration before the surgery, reducing the patient discomfort to a minimum.