Quantitative experimental comparison of HRRT versus HR+ PET brain studies

The High Resolution Research Tomograph (HRRT) is a dedicated 3D brain positron emission tomograph (PET), designed for a resolution of 3 mm or less. Recent improvements in image reconstruction strategies, such as the implementation of ordinary Poisson OSEM, improved the quantitative accuracy of HRRT PET. Further improvement of its accuracy might be expected using a new randoms estimation method based on coincidence histograms. The purpose of the present study was therefore to further evaluate the accuracy of HRRT studies using these new reconstruction methods. Moreover, data will be compared with those measured on a clinical HR+ PET scanner (Siemens), which has been used most frequently for human brain applications so far. To this end, a number of phantom experiment using e.g. NEMA scatter and attenuation, homogeneous (linearity and uniformity) and anthropomorphic brain phantoms, have been performed on both scanners. When using the new randoms estimation method, HRRT experiments showed a remnant scatter fraction < 5%, uniformity < 5% and linearity < 3% up to 136 MBq. These results were similar of better than those obtained with the HR+. HRRT brain phantom studies provided accurate results within 10 and 15% for grey and white matter areas, respectively, for high statistics (> 1800s) scans only. Large under- and overestimations of 20 and 50% in grey and white matter areas, respectively, were observed in case of short acquisition frames (10-30s). As short acquisition frames of 10-30s are normally applied in dynamic brain studies, it is concluded that further refinement of image reconstruction strategies [1] is required to obtain more accurate results, which are comparable with those of the HR+, for dynamic HRRT PET brain studies.