Clustering for three dimensional Kinetic PET Data

Clustering for three dimensional kinetic positron emission tomography (PET) data is considered. Recently, a fast preprocessing clustering technique for kinetic PET data was introduced by Guo et al. 2003. It is, however, still limited with respect to efficiency for three dimensional PET data. Here we present a two level clustering process which combines a slice by slice two dimensional clustering and a classic hierarchical clustering. We compare this method with our previous algorithm by clustering FDG-PET brain data of 12 healthy subjects. The new method significantly reduces the overall time and memory demand without loss of quality. October 15, 2003