Implicit neural representation-based method for metal-induced beam hardening artifact reduction in X-ray CT imaging

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Metal-induced beam hardening artifacts degrade X-ray CT image quality.
• Existing correction methods require precise material segmentation and parameter estimation.
• These limitations hinder artifact correction in complex scenarios involving metals, bone, and teeth.

Purpose of the Study:

• Develop a parameter-free metal-induced beam hardening correction (MBHC) method.
• Overcome limitations of existing MBHC approaches by eliminating segmentation and parameter estimation.
• Provide a generalized solution for metal artifact reduction in CT.

Main Methods:

• Utilize implicit neural representations (INR) to generate monochromatic attenuation and nonlinear beam hardening images.
• Employ a loss function to model predicted projection data nonlinearly.
• Eliminate the need for geometric and parameter estimation of metals.

Main Results:

• Effectively reduced beam hardening artifacts from interactions between metals, bone, and teeth.
• Demonstrated potential in addressing data insufficiency challenges like photon starvation and truncated fields of view.
• Achieved high-quality image reconstructions in numerical and phantom experiments.

Conclusions:

• The proposed generalized MBHC method offers robust artifact reduction without parameter estimation or segmentation.
• This INR-based approach provides a significant advancement for metal artifact correction in CT imaging.
• Enables more accurate diagnoses by improving image quality in the presence of metallic implants or dense materials.