Patient-specific brain phantom SPECT database of cerebral perfusion analysed with the 3D-data analysis platform Scenium

Introduction: Database comparison software, like Scenium, has been developed to aid in the assessment and the quantification of pathologies in the daily clinical routine. Matching of patient data acquired on PET and SPECT scanners with a reference database enables automated analysis through comparison of mean pixel values. Differences between activity ratios reported in the database and measured uptake rations for different regions of the brain help in the evaluation of the PET or SPECT data. In this study, SPECT/CT images of two high-resolution, patient-specific, fillable 3D-printed phantoms were acquired and compared with the Scenium database. These two brain phantoms primarily serve as a ground truth for analysis and correction of the partial volume effect.

Methods: To create these phantoms based on real patient data, a workflow is developed. First, high-resolution patient MRI images are segmented using a semi-automatic segmentation method, followed by the 3D printing of the hull of the 3D brain segmentation, which facilitates filling the phantoms. After printing each phantom in several pieces, the printed phantom was combined using glue and coated with resin to ensure water tightness. For data acquisition, the phantoms were filled with Tc-99m. A ratio of activity between grey and white matter of 2:1 was used, following acquisition of the phantoms on a Siemens Intevo Bold SPECT/CT. Data was reconstructed using Flash3D SPECT reconstruction. The parameters for the reconstruction were chosen similarly to those suggested for the Scenium database comparison. The reconstructions were loaded into the Scenium tool and were registered to match the images of the database. Consecutively, differences between the database and the phantoms were analysed. Scenium not only offers the capability to analyse and compare the uptake of the whole brain, but also displays the difference in uptake of different brain regions. Both evaluation methods are used in the following.

Results: Visual analysis revealed reasonable uptake distributions in the phantoms, although the mean activity of the whole phantom data was 5.4 σ lower compared to the database. In several areas of the phantoms, e.g., the brain stem or the occipital lobe, the observed activity was much lower than in the database, which could be explained by excess resin preventing the Tc-99m solution from reaching these areas. Furthermore, the anatomy of the phantom is based on patient data, meaning that pathologies such as Alzheimer can lead to changes in the brain, causing true differences in activity compared to the database. The 3D segmentation of the patient data can be another source of error for the observed differences in activity. An area for which this effect was observed was the basal ganglia, where too little activity was measured compared to the database.

Conclusions: The results obtained with the help of the Scenium database demonstrate the feasibility of printing 3D phantoms based on brain segmentations, as well as the current limitations of this approach. The results suggest improvement both in the preparation and in the post-processing of the creation of phantoms. During preparation, further steps are necessary to optimize the segmentation. These should concern, in particular, the assembly of the individual segments and their coating with resin, ensuring an even distribution of this substance throughout the phantom, avoiding artifacts in the images engendered later on. Overall, however, it can be seen that good results have already been achieved, on which further research can be built. The next goal is to create more brain phantoms and to minimize the mentioned sources of error.

• Download figure
• Open in new tab
• Download powerpoint