Clinical Applications for a Digital Twin of a Long Axial Field of View PET/CT Scanner

Ziel/Aim: Despite long axial field of view PET/CT scanners leverage innovative applications, inherent limitations of PET imaging, such as motion artefacts or the lack of ground truth information on the absorbed dose, remain. Therefore, our group had developed an in-silico replica of the Siemens Biograph Vision Quadra PET/CT scanner to obtain simulation-derived ground truth to advance various correction methods and dose estimation.

Methodik/Methods: The GATE based Monte Carlo simulation and image reconstruction framework, closely resembling the event processing and image reconstruction of the system, was validated against real experiments following the NEMA NU 2-2018 standard. Furthermore, this digital twin was used to assess the impact of motion and motion correction (Siemens Oncofreeze AI) with the digital patient-like XCAT phantom, modelling respiratory motion, and the impact on lesion (lung, liver; diameter 5-20 mm) blurring and quantification.

Ergebnisse/Results: The comparison with experiments verified high accuracy in replicating image quality, contrast recovery (IEC phantom 17 mm sphere: 74.9% | 74.6% (simulation | experiment)), image noise, sensitivity (82.7 cps/kBq | 82.6 cps/kBq (simulation | experiment)), spatial resolution and count rate. Compared to a motion free case, the motion induced lesion quantification error was -41% (5 mm) and -6% (20 mm) and was mitigated by motion correction to -28% (5 mm) and -3% (20 mm).

Schlussfolgerungen/Conclusions: Following the NEMA NU 2-2018 validation, we showcased, as an example application, the potential of the digital twin to quantify motion artefacts and benchmark different motion correction methods. Our current efforts extend this framework with realistic patient models derived from Y-90 imaging data of patients scanned post radioembolization therapy. Here, patient-specific Monte Carlo simulations of the absorbed dose serve as ground truth to study the effects of variations in image reconstruction, motion/scatter correction and dose estimation method on dose maps – aiming to advance accuracy and facilitate clinical implementation of dosimetry. In addition, our group currently uses this framework to study the effect of not accurately accounting for event rates when reframing high-dose image data to train neural networks for low-dose PET image denoising.