Quantitative Evaluation of three Siemens Biograph PET scanners using different reconstruction algorithms and parameters.

Purpose: Our department is equipped with 3 PET/CT scanners of different generations and in order to match the resolution and noise properties of these scanners, this study compared their performance using different reconstruction algorithms and parameters, evaluated count recovery and noise, and determined image reconstruction parameters leading to harmonized quantitative imaging performance among all the three scanners.

Methods: The NEMA-IEC image quality phantom with the standard 6 spheres (10 to 37 mm) and a second set of 6 smaller spheres (4-12mm) were filled with 9.35:1 lesion to background ratio. Measurements were first conducted with the Siemens Vision PET/CT, the Biograph mMR PET/MR and the Biograph 40 PET/CT. Data were acquired in list-mode format, for the Vision PET/CT images were generated at 1 and 5 minutes frame duration. For the successive measurements, due to the radioactive decay, this time frames were extended have a match in the activity for all three scanners: time frames were extended to 1.23 and 6.33 minutes for the B40, and to 1.62 and 8.42 minutes on mMR. Images were reconstructed using 3D-ordered subset expectation Maximization (3D-OSEM) algorithm with and without including point spread function (PSF). Time of flight (TOF) was employed on the Biograph Vision. The number of iterations were varied from 4-8 on the Vision, and from 2-5 on the B40 and the mMR. Images were reconstructed with a range of 0-5 mm Gaussian post-reconstruction smoothing filter. The contrast recovery coefficients (CRC - ratio of mean measured activity within a sphere over the expected activity) and coefficients of the variance (COV - standard deviation of the pixel intensity in a large VOI in a uniform section of the phantom divided by the mean value within this VOI) were used as metrics of count recovery and image noise for this study. Lesion detectability with the smallest set of spheres was conducted.

Results: Using the standard set of spheres, contrast recovery curves were created for each scanners and each image reconstruction parameter sets (filter and iteration number). The configuration of image reconstruction parameters that led to the most identical count recovery curve among the three scanners was determined from the sum squared difference minimized over all image reconstruction parameter set. This study indicated that image reconstruction performed with 4 iterations, with respectively 4, 0 and 5 mm post-reconstruction filter on the B40, mMR and Vision scanners (5min data) led to harmonized CRC curves (Fig1a). When using point-spread function, the best agreement was obtained at 4it/5mm on B40, 2it/0mm on mMR and 4it/4mm on the Vision with the Vision producing still higher CRC value for all spheres (Fig1b). Best agreement with the 1 minute of data (more noisy) was obtained with 5it/2mm on B40, 2it/0mm on mMR and 4it/5mm on the Vision. The CRC vs COV analysis indicated that PSF reconstruction leads to increased recovery without significant noise increase (Fig 2a, 2b). At matched activity, the COV in the Vision is overall lower (lower noise). Lesion detect ability indicated the Vision scanner was most superior to identify small lesions.

Conclusions: This study allowed to fully map the range of size-dependent CRC values from three PET/CT scanners in our department and to determine image reconstruction parameters that permit to obtain quantitatively comparable PET data from these three scanners. This study also demonstrate the improved performance of the Vision for identifying and quantifying small lesions.

• Download figure
• Open in new tab
• Download powerpoint