Abstract
We present a fast Hierarchical linkage (HL) clustering technique for
kinetic positron emission tomography (PET) data according to a dissimilarity
measurement between time activity curves. Due to the huge size of PET data,
the cost of conventional HL is too expensive to real applications. We suggest
two strategies: significantly reduce the data size, about 75%, by
thresholding and a preclustering technique and followed by a slice by slice
two dimensional HL clustering. The efficiency and superiority is validated
by comparison of the results for FDG-PET brain data of 13 healthy subjects.
This is joint work with R. Renaut and K. Chen
