Performance Characteristics of Iterative Image Reconstruction Techniques for Routine Use in Positron Emission Tomography.

Introduction

Image reconstruction is one of the basic tasks in nuclear medicine to achieve a high image quality. The filtered backprojection (FBP) algorithm is one of the most commonly used procedures for the reconstruction of cross sections. The FBP is well known in computed tomography (CT) and provides generally an acceptable image quality. In contrast to radiological procedures, nuclear medicine studies deal with less information per projection as compared to radiological procedures like CT. Therefore, the FBP may result in images with limited quality. While the FBP algorithm is fast and images are obtained with short reconstruction times, limitations exist with respect to image quality, especially when high regional activity concentrations are present or in studies with low count rates.

Besides the FBP technique, other approaches were applied for image reconstruction.Iterative image reconstruction techniques were introduced more than 10 years ago and have been found useful especially for positron emission tomography (1-2). However, limitations exist for the routine application of this technique due to the higher computational demand and the slow convergence of the algorithm. Therefore, the iterative reconstruction was mainly limited to major workstations and several attempts, including the implementation on parallel computer systems, were made to speed up the reconstruction task (3-5). Due to the availability of new, powerful PC systems, the iterative reconstruction can now be implemented on PC and used for routine patient studies. Furthermore, clustering or "semi-parallel"-processing of the reconstruction data may help to enhance the reconstruction process. We have recently published the design and implementation of a new PC based iterative reconstruction program, which is in use at our center for PET studies (6). The performance characteristics of the iterative reconstruction and the evaluation of the different reconstruction algorithms and parameters are presented in this paper. While phantom studies were already performed to evaluate the basic properties of the iterative reconstruction, we noted limitations when the results from phantom studies were compared to those obtained in patients. Respiration movement is usually present in PET studies of the whole body area and deteriorate the imageresolution. Furthermore, the effect of attenuation due to very different tissue structures and the influence of the different shapes of lesions as well as organs is difficult to simulate with phantoms. Therefore, a typical dynamic patient study was used to assess the reconstruction parameters which are important for PET patient studies.