Author: Naruomi Akino, Kirsten Lee Boedeker, Ilmar Hein, Akira Nishikori, Daniel W Shin 👨🔬
Affiliation: Canon Medical Systems Corporation, Canon Medical Research USA 🌍
Purpose: To validate a simulation tool using physics-based image quality metrics in both phantom and patient data, and to assess the low contrast detectability (LCD) of Super Resolution-Deep Learning Reconstruction (SR-DLR) compared to conventional reconstruction algorithms through an in-silico imaging trial.
Methods: An analytic forward projection model based on the specifications of a wide-volume CT scanner (Canon Medical Systems, Aquilion ONE PRISM, Otawara, Japan) including detector and source geometry with beam spectra, was used for simulation. Validation was conducted using the CATPHAN™ 500 and water phantoms (24 cm and 32 cm diameter) scanned at various mA levels and reconstructed with filtered backprojection (FBP). Simulated and acquired images of the phantoms were compared for CT number, noise power spectrum (NPS), and modulation transfer function (MTF). The MTF of simulated CATPHAN™ sensitometry cylinders within an anatomical background was compared to the MTF obtained from the actual CATPHAN™. For the LCD study, low contrast rods were simulated and inserted into the liver region of a patient sinogram. The sinograms were reconstructed using SR-DLR, Hybrid Iterative Reconstruction (HIR), and FBP. Detectability (d’) was assessed using a non-prewhitening model observer, with standard error estimated via bootstrapping.
Results: The CT numbers, MTF, and NPS measured from the simulated images closely matched those of actual CT system for both phantom and patient data, confirming the simulation tool's validity. In the in-silico LCD analysis, SR-DLR demonstrated significantly improved detectability compared to HIR and FBP. SR-DLR showed 24–33% and 109–131% higher detectability than HIR and FBP, respectively, with all comparisons yielding p-values < 0.01.
Conclusion: The validated simulation tool reliably generates images equivalent to those from an actual CT system, enabling rigorous in-silico assessments of DLR. SR-DLR demonstrated superior LCD compared to conventional reconstructions, supporting its potential for improved diagnostic confidence and dose optimization in clinical practice.