Formation of parametric images in positron emission tomography using a clustering-based kinetic analysis with statistical clustering

A method is proposed for forming parametric images in positron emission tomography, using clustering kinetic analysis. To overcome the dual problems experienced in voxel-based data, of signal noise and the very long computational time, the data are clustered before parameter estimation, and then an estimation procedure is applied to the averaged data in each cluster. Using this algorithm, PET data are optimally clustered, depending on the noise that is present, by hierarchically applying a statistical-clustering algorithm based on mixed Gaussian model. In a computer simulation, the proposed method correctly clustered noise-contaminated data. Applying the proposed algorithm to ¹⁸F-FDG clinical data, physiologically acceptable parametric images of glucose metabolism in a brain were obtained in a practical calculation time.