No abstract

Detection of parathyroid tumor using conventional imaging modalities such as Sestamibi and 4D CT suffer from poor resolution or excessive radiation to the parathyroids. Dynamic Contrast Enhanced MRI (DCE-MRI) is emerging as a viable option for detecting parathyroid tumors. However, conventional quantitative methods to segment tumors from DCE-MRI, which include black-box methods and pharmacokinetic models, are highly sensitive to imaging noise, inhomogeneity, timing of the contrast injection and image acquisition. Time series analysis has proven to be a useful tool to extract features from the data in the presence of noise and signal uncertainty. In this paper, we model the underlying tissue as a Linear Dynamic System (LDS) and estimate the system parameters using the timeintensity curves observed at each voxel. The system parameters are then clustered into healthy and tumor class. The result of the LDS based segmentation algorithm, compared to the radiologist's segmentation, shows accurate delineation of the tumor and robustness to imaging noise.