A Simple Adaptive Bandwidth Scheme that Improves Image Quality and Kinetic Quantitation in Dynamic PET Scans

Objectives: Dynamic positron emission tomography (PET) scans give the ability to recover detailed tracer kinetic information about tissue. While a fixed resolution filter bandwidth is typically used in the reconstruction of all dynamic time-frames. This may over/under smooth some high/low counts time frames. We propose a simple procedure that adaptively modifies time-frame bandwidths in order to improve the quantitative accuracy of dynamic PET images and derived kinetic information.

Methods: The adaptive procedure assumes that an overall fixed bandwidth for the reconstruction resolution filter, suitable for the total tracer uptake, has been specified. Using established asymptotic analysis of reconstruction error as a function of dose, the fixed bandwidth is manipulated by time-frame based on the observed counts in the frame. Performance of the adaptive/dynamic and fixed bandwidth is evaluated in using a series of simulation studies matched to standard dynamic sampling schemes for O-15 water (H2O) and F-18 Fluorothymidine (FLT) studies. Using literature data, field of view tissue structures corresponding to what one would typically encounter in PET breast cancer studies with these tracers are specified. A simplified scanning model matched to the mathematical complexity of PET is used. To facilitate accurate estimation of performance, simulations are conducted at several doses, with a large number of replicates at each dose. Quantitative kinetic analysis voxel-level metabolic parameters including flux (Ki), volume of distribution (Vd) and flow (K1) are computed using both standard fixed-bandwidth dynamic data and also the dynamic data derived from the adaptive bandwidth procedure. The relationship between root mean square error (RMSE) characteristics of the image quality and local kinetic parameters for these two schemes are described.

Results: A simple and highly efficient implementation in the technique is described - an open-source R coding is provided. RMSE comparisons of reconstructed images and metabolic parameters across different dose levels from fixed and dynamic bandwidth approaches are presented in terms of the percent (Figure). These results show that the dynamic scheme substantially improves reconstructed image quality - percent improvements are on the order of 25% across a range of doses. Improvements in metabolic parameters are not as substantial but still significant - typically on the order of 10%. For flux and volume of distribution, RMSE improvements range from 7% to 15%; for flow, improvements range from 6% to 32%. Conclusions: The superiority of the adaptive bandwidth scheme are fully demonstrated in simulation studies. This scheme has the potential to be applied to smooth the dynamic PET images and increase the accuracy of derived kinetic information.

• Download figure
• Open in new tab
• Download powerpoint