A rapid and accurate method to realign PET scans utilizing image edge information.

UNLABELLED: Movement during or between PET examinations is a common and serious problem. Consequently, there is a great need for rapid, accurate and robust methods to realign image sets.
METHODS: Derivative information from the image sets was used to extract areas containing edge information. Image similarity between a reference dataset and a misaligned dataset was evaluated for these areas. Powell's method for function minimization was used to find the set of translations and rotations along and around the axes that maximized image similarity. The method was validated by realigning image sets with a known misalignment. Image sets used for validation included brain studies using several different tracers and heart studies using labeled acetate or water.
RESULTS: The method was capable of labeled acetate or water.
RESULTS: The method was capable of realigning brain datasets using the same tracer with an accuracy of 0.2 mm and 0.2 degrees along and around all axes. The same accuracy was obtained for datasets with as few as a total of 800,000 counts. Brain studies utilizing different tracers with markedly dissimilar regional uptake patterns were realigned with an accuracy of 1.5 mm and 1.5 degrees. Heart studies using water or acetate were realigned with an accuracy of 0.2 mm and 0.4 degrees along and around all axes. Realignment of a heart study containing a large focal uptake defect against a dataset without defect produced errors no greater than 1.0 mm and 1.0 degree.
CONCLUSION: The use of derivative information provides a useful method to accurately realign PET image sets. It is rapid and noise-insensitive enough to allow for its routine use in dynamic studies.