Material depth reconstruction method of multi-energy X-ray images using neural network

Area of Science:

• Medical Imaging
• Computational Physics
• Artificial Intelligence

Background:

• Multi-energy X-ray imaging offers reduced patient dose and functional imaging capabilities.
• Photon-counting detectors are crucial for developing advanced multi-energy imaging systems.
• Material decomposition is key to extracting quantitative information from multi-energy X-ray data.

Purpose of the Study:

• To present a novel material decomposition method for multi-energy X-ray images.
• To utilize a feedforward neural network for accurate material depth reconstruction.
• To demonstrate the efficacy of the proposed method using simulated phantom data.

Main Methods:

• Monte Carlo simulations were employed to generate multi-energy X-ray images, incorporating spectrum modeling and ripple effects.
• Material decomposition was achieved by leveraging energy-dependent X-ray attenuation differences.
• A feedforward neural network was trained using step wedge phantom images to map multi-energy data to material depth.

Main Results:

• The neural network successfully decomposed simulated multi-energy X-ray images into material depth images.
• The method demonstrated effective material depth reconstruction for a 3D head phantom.
• The approach shows significant potential for quantitative material analysis in medical imaging.

Conclusions:

• The developed neural network-based material decomposition method is effective for multi-energy X-ray imaging.
• This technique can accurately reconstruct material depth, offering functional imaging insights.
• The findings support the advancement of photon-counting detector-based multi-energy X-ray imaging applications.