Positron emission tomography (PET) is nuclear medicine imaging technique that produces a three-dimensional image of functional processes in the body. The system detects pairs of gamma rays emitted by a tracer, which is introduced into the body. Three-dimensional images of tracer concentration within the body are then constructed by computer analysis. Respiratory motion in emission tomography leads to image blurring especially in the lower thorax and the upper abdomen, influencing this way the quantitative accuracy of PET measurements as well a leading to a loss of sensitivity in lesion detection. Although PET exams are getting shorter thanks to the improvement of scanner sensitivity, the current 2-3 minutes acquisitions per bed position are not yet compatible with patient breath-holding. Performing accurate respiratory motion correction without impairing the standard clinical protocol, ie without increasing the acquisition time, thus remains challenging. Different types of respiratory motion correction approaches have been proposed, mostly based on the use of non-rigid deformation fields either applied to the gated PET images or integrated during an iterative reconstruction algorithm. Evaluation of theses methods has been mainly focusing on the quantification and localization accuracy of small lesions, but their impact on the clinician detection performance during the diagnostic task has not been fully investigated yet. The purpose of this study is to address this question based on a computer assisted detection study. We evaluate the influence of two motion correction methods on the detection of small lesions in human oncology FDG PET images. This study is based on a series of realistic simulated whole-body FDG images based on the XCAT model. Detection performance is evaluated with a computer-aided detection system that we are developing for whole-body PET/CT images. Detection performances achieved with these two correction methods are compared with those achieved without correction, ie. with respiration average PET images as well as with reference images that do not model respiration effects. The use of simulated data makes possible the creation of theses perfectly corrected images and the definition of known lesions locations that serve as a reference.
