A Neural-Analytical Fusion Scatter Correction Method for Multi-Source CT Using Equivalent High-Order Scatter

Area of Science:

• Medical Imaging
• Computational Physics
• Radiological Sciences

Background:

• Multi-source computed tomography (MSCT) offers improved temporal resolution but is prone to significant scatter artifacts.
• Existing scatter correction methods, including model-based and deep learning approaches, have limitations in accuracy and physical constraint adherence.
• Addressing scatter is crucial for enhancing diagnostic accuracy in MSCT.

Purpose of the Study:

• To develop and validate a novel neural-analytical fusion (NAF) scatter correction method for MSCT.
• To improve the accuracy and physical consistency of scatter correction, particularly for high-order scatter.
• To reduce scatter artifacts without additional hardware or radiation dose.

Main Methods:

• Developed a NAF method combining analytical estimation of first-order scatter (Compton and Rayleigh) with a deep learning approach for high-order scatter.
• Integrated an equivalent high-order cross-section prediction network (EHCP-Net) within the analytical model for physically constrained estimation.
• Validated the method on simulated and real MSCT data across various scanning geometries using GPU acceleration.

Main Results:

• The NAF method demonstrated superior scatter correction accuracy compared to state-of-the-art techniques.
• Achieved a mean absolute percentage error (MAPE) < 3% for scatter distribution compared to Monte Carlo simulations.
• Yielded mean absolute errors (MAE) < 20 HU on simulated data and < 30 HU on real data for scatter correction.

Conclusions:

• The proposed NAF method effectively corrects scatter artifacts in MSCT with strong physical constraints.
• This approach enhances image quality by suppressing artifacts and improving accuracy.
• NAF offers a promising software-based solution for scatter correction in MSCT imaging.