Optimization and evaluation of landmark-based image correlation

Abstract
Image correlation methods enable the complementary use of information from different medical images of a patient. These images can be obtained from different imaging devices (CT, MR, PET), or, from one imaging device taken at different times. Unfortunately, there are few cases in which the requirements for later image correlation are taken into account at the time of image acquisition. There is therefore a need for correlation techniques requiring no preparation in advance. We have developed two correlation methods, both based on three or more anatomical or artificial landmarks, to be defined in corresponding image data sets. These methods have been evaluated with phantom data as well as with patient data. We have improved these correlation methods by using more landmarks and special selection criteria. They are applicable to all medical tomograms and to x-ray pictures taken under stereotactical conditions. The results obtained have error ranges in the order of the three-dimensional image resolution.