Our goal is to determine the extent to which state-of-the-art acquisition and reconstruction strategies can overcome the factors that limit the detection of, and estimation of activity within, tumors in gamma-camera-based emission-computed-tomographic imaging of tumor-avid imaging agents. Altering acquisition and reconstruction strategies to account for a source of degradation is costly in terms of processing time, added complexity in imaging and processing, and/or enhancement of other sources of degradation. Therefore, it is essential to establish the degree to which degradations are detrimental, and to determine the extent to which their impact can be mitigated. Thus far, we have determined using human-observer LROC studies employing simulated Ga-67 citrate images that noise regularization, nonuniform attenuation compensation (AC), and detector resolution compensation (DRC) significantly increase the detection accuracy of small, low-contrast lesions. We propose to proceed by investigating scatter compensation (SC), and the explicit inclusion of noise regularization in iterative reconstruction (IR). We also propose to investigate the impact of respiratory motion on tumor detection. These studies will be conducted with simulations where imaging can be accurately modeled and upper limits of improvement assessed. However, simulated images do not fully reflect the structured background present in clinical images, and are created with system models that only approximate actual systems. Therefore, we propose to investigate noise regularization, AC, DRC, and SC through hybrid images, which are actual clinical images to which Monte-Carlo-simulated tumors have been added. To investigate the generalizability of our findings, we will expand the clinical studies using hybrid images beyond solely Ga-67 citrate imaging of lymphoma to include Tc-99m-labeled NeoTect imaging of solitary lung nodules (SLN), and FDG coincidence imaging of SLN and mediastinal lesions. The regularization parameters associated with each strategy will be optimized for maximal detection performance through human-observer LROC studies. Numerical observers will be employed to narrow the parameter space considered by human observers. We will also investigate new formulations of numerical observers with the goal of further improving their ability to predict the relative rankings of human observers for imaging strategies. As indicated by the title of this proposal, we emphasize tumor detection in the investigations; however, due to its potential clinical importance. we will also investigate tumor-activity quantitation.
Showing the most recent 10 out of 14 publications
