Incorporating patient-specific variability in the oncoPET_DB database

The purpose of this study was to generate a series of simulated 18F-FDG PET oncology images incorporating variability of patient anatomy and 18F-FDG uptake in the main organs as well as realistic respiratory motion. These data are based on the XCAT model adapted to a series of clinical CT data and on patient respiratory cycles derived from spirometry records. Spherical 8 and 12 mm diameter lesions of homogeneous uptake were randomly distributed in the lungs and the liver, which are both significantly impacted by respiratory motion. Five different lesion contrasts per diameter and per organ were sampled based on a human observer detection study so as to cover the entire range of detectability. The simulated database was generated with the PET-SORTEO Monte-Carlo simulation tool that was fully validated against the geometry of the GEMINI system (Philips Healthcare, Cleveland, OH). The simulated database is actually composed of 15 whole body PET simulated images with a total of 107 tumors in the liver and 173 tumors in the lungs. These data will complete the first series of images of the oncoPET_DB database (www.creatis.insa-lyon.fr/oncoPET_DB) by providing the raw list-mode files of the 15 models as well as four sets of reconstruction per model. These four series include the extreme cases of respiration-free (non breathing) and free-respiration (no correction) images and two respiratory motion corrected images based on the estimation of elastic transformation fields applied either on the reconstruction system matrix, during reconstruction, or a posteriori, on the reconstructed PET gated images. Perspectives include performing a human FROC study on the 15 images to validate the lesion contrast calibration.