Automatic Descending Aorta Segmentation in Whole-Body PET-CT Studies for PERCIST-Based Thresholding

摘要

Medical imaging is a fundamental component of modern healthcare where majority of medical conditions can benefit from some kinds of imaging. A dual-modal positron emission tomography - computed tomography (PET-CT) has increasingly become the preferred imaging method to stage the common cancers and to assess treatment response. For example, PET image can quantitatively assess the treatment before morphological changes that can be detected on anatomical CT image. PET response criterion (PERCIST) is a widely recognised thresholding method for detecting malignant lesions (high metabolic value) in treatment response. It is based on the calculation of standardised uptake value with lean body mass (SUVLBM), together with a volume of interest (VOI) reference placed on the right lobe of the liver or the descending aorta (when the liver is abnormal e.g., liver cancer). These two structures are considered to have stable metabolism among the PET patient population and therefore can be used to normalise the SUV. The current practice of PERCIST thresholding depends on the manual delineation of the VOI reference which is a time consuming and operator-dependent process. Furthermore, such VOI selection is more difficult on the smaller descending aorta structure when compared to the liver. In this study, we propose a fully automatic approach to the segmentation of the descending aorta for use in the calculation of the PERCIST thresholding. A multi-atlas registration coupled with weighted decision function was used in the whole-body CT for the segmentation of the descending aorta, with the resulting VOI reference mapped to the co-aligned PET counterpart. We evaluated our method with 30 clinical PET-CT studies with preliminary results demonstrating reliability and robustness.