Investigation of lesion detectability in dynamic PET data sets

Dynamic positron emission tomography (PET) has strong potentials for improving detection of cancers at their early stage. However, dynamic PET images are difficult to manipulate by human observers in the detection task because of the huge data size (313 in space + ID in time). One approach to reducing the size is to fit the dynamic images into a compartmental model and use the kinetic parameter images for tumor detection. In this paper we use numerical observers to investigate the performances of different approaches to the utilization of temporal information in dynamic PET for the detection task. Both object variability and measurement noise are considered in the Monte Carlo simulation. The results provide a useful guidance for the proper utilization of dynamic PET for tumor detection.