Effects of SPECT collimation and system geometry on classification tasks related to Parkinson's disease

The authors assessed the potential performance of three commercial SPECT systems in simulated, but realistic, imaging tasks related to the diagnosis and management of Parkinson's disease (PD). Images of I-123 altropane activity distributions in normal and PD brains were modeled assuming values from the literature for striatal sizes and activity concentrations, as well as for nonspecific activity. Imaging characteristics of the three systems were determined by phantom studies. The expected distributions of estimates of activity concentration and striatal volume were based on covariance matrices determined by the Cramer-Rao lower bounds calculated for a simultaneous estimation of these two parameters, as well as background activity concentration. Potential performance in several binary classification tasks was assessed by fitting ROC curves to simulated likelihood ratios for 5000 'subjects' in each category. For all conditions studied, ROC areas were similar for the higher-resolution instruments and lower for the lower-resolution system. These results imply a diagnostic advantage for higher-resolution systems in imaging tasks related to PD.