LuCaS: Efficient Monte Carlo simulations of highly realistic PET tumor images

Abstract

In this work, we propose a method to perform highly realistic and efficient simulations of fluoro-deoxy-glucose (FDG) positron emission tomography (PET) acquisitions of patients with lung cancer. Using patient PET images, contours of lung tumors were manually delineated by a nuclear physician. Using these tumor shapes, Monte Carlo (MC) simulations of PET acquisitions of the tumors only, including uniform or heterogeneous activity distributions, were performed with GATE (Geant4 application for emission tomography), by imbedding the tumors in an attenuation medium derived from the computerized tomography (CT) scan of a healthy subject. Each tumor sinogram was merged with the sinogram of the PET acquisition of the healthy subject. The reconstruction of the merged sinograms yielded realistic tumor images including all physiological heterogeneities of the tracer distribution in non-tumor tissues. Using such simulated images, we showed that the performances of an algorithm for tumor segmentation could be far too optimistic when assessed from a simple phantom with spheres compared to what they actually are in more various and realistic simulated configurations. The proposed simulation approach, by modelling only the tumor activity using the MC approach, is about 100 times faster than a complete simulation of an anthropomorphic phantom, and can be used to generate large datasets for evaluation purpose.