In silico modeling of a clinical photon-counting CT system: Verification and validation

Computed tomography (CT) is rapidly advancing with the recent FDA clearance of photon-counting CT (PCCT) systems for clinical use. New technologies are best evaluated using physical phantoms or patient images. However, both methods have their drawbacks: physical phantoms lack the complexity and diversity to be representative of a human population, and patient images lack the ground truth and are often cost- and time-prohibitive. Virtual imaging trials, also known as in silico trials, can mitigate these limitations by using computer simulations to conduct imaging experiments to assess the safety and efficacy of new medical technologies for specific clinical conditions. However, to date, there have been no simulation platforms to assess PCCT systems generalizable across clinical imaging conditions.

The purpose of this study was to develop and validate a computationally-efficient CT simulator that realistically accounts for conditions of a clinical PCCT system to be used for in silico trials.

The simulator was built upon a state-of-the-art platform (DukeSim) which simulates CT projection images of computational phantoms given the geometry and physics of the scanner and acquisition settings. To model the photon-counting detection process, we utilized a spatio-energetic detector model based on Monte Carlo simulations with known properties of the detectors. DukeSim was augmented to account for the geometry and physics of a clinical PCCT system (NAEOTOM Alpha, Siemens), including correlated noise between detector thresholds, and low-signal physics for photon starved regions. We validated the simulation platform against experimental measurements using two physical phantoms scanned with a clinical PCCT scanner. The ACR phantom-a cylindrical phantom with four inserts was used to validate the simulator across a clinically relevant dose range with images acquired at four dose levels (CTDI_vol of 1.5-12.0 mGy). The Mercury phantom-a cylindrical phantom with five diameters (160-360 mm) was used to validate the simulator for a range of patient sizes. The experimental acquisitions were replicated using our developed simulation platform. Each acquisitions was reconstructed using ReconCT with three kernels at a 0.4 mm slice thickness. The real and simulated images were quantitatively compared in terms of image quality metrics. Further, the simulation spatial resolution uncertainty from input parameters was quantified in terms of the MTF.

The developed simulator generated images that were close to the experimental scans. In terms of noise magnitude, the discrepancy between real and simulated data were 2.78 ± 3.02% and 2.67 ± 1.53%, for the ACR and Mercury phantom, respectively. In terms of the frequency at 50% MTF, the discrepancy between real and simulated data were 0.05 and 0.03 mm^-1, for the ACR and Mercury phantom, respectively. Subsampling had a strong effect on the MTF measurement ( ${\Delta }_{Euclideandistance}$  = [0.25, 1.12]), while voxel size had a weaker effect ( ${\Delta }_{Euclideandistance}$  = [0.48, 0.55]).

We successfully developed and validated a virtual imaging platform for a current PCCT system, enabling the optimization of imaging parameters (including dose optimization) for specific clinical tasks, as well as comparisons across CT systems.