Automatic motion correction in dynamic radionuclide renography using image registration

Abstract

Time-activity curves from dynamic renograms can be analysed to yield important quantitative parameters. However, accurate results are dependent on curves being free from artefacts caused by patient movement. We describe a method to correct for both translational and rotational motion during renography. The method is image-based and does not require markers. Sequential dynamic frames are registered using an affine transform from which rotational and translational components are extracted by singular-value decomposition. Computer simulations demonstrate correction to < 1 pixel and < 1 degree. The method is also fast, with renograms comprising 75 frames of 64 x 64 pixels taking under 1 min for correction on a Sun SPARC 5.