Dynamic 4D PET Reconstruction Using the Spectral Model and Adaptive Residual Modelling

Application of 4D reconstruction algorithms to dynamic data, especially in whole-body dynamic imaging, can result in spatial propagation of errors over the field of view originating from regions with poor model fits. An adaptive modelling strategy has been previously proposed for this problem to improve the model fit over the field of view, using residual adaptive modelling in the reconstruction process. We used this strategy within a 4D reconstruction algorithm that makes use of the spectral analysis model as a primary model and a PCA based adaptive residual model as a secondary model. The developed algorithm maintains genericity and does not impose strong assumptions about the underlying kinetics. The objective of this work is to evaluate the algorithm on a whole-body dynamic dataset of a healthy volunteer imaged with 11C-Glyburide, a newly developed tracer used in the study of distribution and function of OATP transporters. Results showed improved estimates over the pelvic area where the bladder filling process was poorly fitted by the spectral model alone. The adaptive 4D algorithm has successfully identified and selectively corrected for the filling process, when compared against 3D independent frame reconstructions, while preserving the fit of well-modelled regions. Optimal results were obtained when spatial and temporal filtering of residual data was applied prior to PCA analysis, in which case the adaptive algorithm selectively corrected poorly modelled physiological processes, averted from modelling respiratory motion induced errors and resulted in reduced induction of noise in the reconstruction process.