METHODS FOR IMPROVING RELIABILITY OF GLLS FOR PARAMETRIC IMAGE GENERATION

摘要

The computationally efficient generalized linear least square (GLLS) algorithm has been successfully applied in positron emission tomography (PET) for constructing parametric images from dynamic data. Recently, we have applied GLLS to substantially more noisy single photon emission computed tomography (SPECT) data. Due to the nature of the data, GLLS sometimes failed to provide physiologically meaningful estimates. In this work, we investigated two potential methods to improve the reliability and success rate of GLLS for estimating kinetic parameters from noisy data:-incorporation of the volume of distribution (V_d) prior into GLLS and applying the bootstrap Monte Carlo method. Our simulation results show that both methods can improve the parameter estimation reliability at the expense of extra computation time.