Segmentation technique is widely accepted to reduce noise propagation from transmission scanning for positron emission tomography. The conventional routine is to sequentially perform reconstruction and segmentation. A smoothness penalty is also usually used to reduce noise, which can be imposed to both the ML and WLS estimators. In this paper we replace the smoothness penalty by a segmentation penalty that biases the object toward piecewise-homogeneous reconstruction. Two updating algorithms are developed to solve the penalized ML and WLS estimates, which monotonically decrease the cost functions. Experimental results on simulated phantom and real clinical data were both given to demonstrate the effectiveness and efficiency of the algorithms which were proposed.
Keywords: Segmentation penalty; auxiliary function; fuzzy c-means clustering (FCM); space alternating descent.