A Computationally Efficient Method to Evaluate and Optimize Reconstruction Parameters and Data Analysis of PET Images. Y. Ma, O. Rousset, and A.C. Evans. McConnell Brain Imaging Centre, Montreal Neurological Institute, Montreal, Quebec.
Positron emission tomography allows quantitative study of the brain function by providing local concentrations of radioisotopes. These data are usually contaminated by artifacts from tomograph design and image reconstruction algorithms. Specifically, accuracy and precision of the regional measurements depend on radiotracer distribution in the brain, distortion corrections in the scanner, reconstruction parameters and regions of interest in use. In order to optimize data analysis, it is necessary to have a simulation tool to evaluate the effects of these variables on image quality.
We present a computationally efficient method to model the data acquisition and image reconstruction of a PET scanner. This method employs digital brain models created from MR data to represent 3-D distributions of tissue activity and attenuation coefficients in typical imaging protocols. Projection data are computed with measurable characteristics such as 3-D detector response, attenuation, scatter, randoms and Poisson noise. Images are then reconstructed by filtered backprojection with a set of different filters and analyzed with regions of interest of varying shapes and sizes. Validation results with anatomically realistic phantoms demonstrate that the simulation algorithms are reliable, and can be used to reveal quantitative artifacts and optimize PET neurological imaging.