Complementary reconstruction: Improving image quality in dynamic PET studies

One widely used method is the individual frame-by-frame reconstruction (IFR) method where the whole scan time duration is divided into multiple time frames, and the data in the each individual time frame is reconstructed separately. All the reconstructed images are used to obtain the temporal tracer uptake changes, and estimate kinetic parameters for quantitative studies. The method often has a drawback having low accuracy on the estimation process due to limited count rates when the frames are divided into very short time intervals to achieve enough temporal resolution. Forward model based iterative reconstruction methods have a potential to improve image quality, and a lot of research has shown their superior performance compared to analytical reconstruction methods, such as filtered-back projection (FBP). The (OS-)EM iterative reconstruction is now widely used in both pre-clinical and clinical imaging studies, and can be done very fast [2]. However, the non-negativity constraint imposed in such iterative reconstruction algorithms tends to produce asymmetric noise distribution (similar to Poisson distribution) in the reconstructed image. Here, we propose a novel reconstruction method which allows negative pixel values, and changes the noise distribution into a symmetric distribution (similar to Gaussian distribution) while still using iterative reconstruction methods.