A GPU-accelerated framework for rapid estimation of scanner-specific scatter in CT for virtual imaging trials

Area of Science:

• Medical Physics
• Computational Imaging
• Radiological Sciences

Background:

• Virtual imaging trials (VITs) offer a cost-effective alternative to traditional clinical trials for computed tomography (CT).
• Accurate scanner-specific scatter simulation is crucial for the clinical viability of VITs but presents computational challenges.
• Existing methods struggle to simulate scatter realistically and efficiently for various CT configurations.

Purpose of the Study:

• To develop and validate a rapid scatter estimation framework for CT virtual imaging trials.
• To enable accurate and time-efficient simulation of scatter in single-source, dual-source, and photon-counting CT.
• To assess the framework's performance using physical measurements and phantom studies.

Main Methods:

• Developed a CT simulator incorporating parametric models for anti-scatter grids and energy-integrating detectors.
• Employed GPU-accelerated Monte Carlo simulations for scatter estimation.
• Integrated denoising methods (convolutional neural network, optimized Gaussian filter) to manage statistical noise and computational load.

Main Results:

• Validated scatter estimates against physical measurements, showing agreement within 4.4% ± 10.8%.
• Demonstrated the significant impact of scatter on simulated CT images (differences of ~121 HU).
• Achieved rapid scatter simulation (approx. 30 seconds per projection) with preserved scatter distribution accuracy (rRMSE 0.91-0.97).

Conclusions:

• The proposed framework provides a rapid and accurate method for scatter estimation in CT.
• This advancement enhances the realism and efficiency of virtual imaging trials.
• The tool is suitable for extensive VITs involving multiple patients and scan protocols.