Monochromatic image reconstruction via machine learning.

Area of Science:

• Medical Imaging
• Computational Imaging
• Machine Learning in Radiology

Background:

• X-ray computed tomography (CT) is crucial for medical diagnosis, but current methods approximate complex physics, leading to inaccuracies like beam-hardening effects.
• The non-linear Beer-Lambert law accurately describes X-ray imaging but lacks efficient computational solutions, forcing approximations that lose energy-dependent information.

Purpose of the Study:

• To develop a machine learning-based approach for quantitative CT imaging by correcting model mismatch.
• To achieve accurate virtual monochromatic (VM) imaging from polychromatic X-ray data.

Main Methods:

• A one-dimensional neural network was designed to learn a non-linear transform.
• The network maps polychromatic CT images to monochromatic sinograms at a specific energy level.
• This enables efficient and effective virtual monochromatic (VM) imaging.

Main Results:

• The proposed method successfully recovers high-quality monochromatic projections.
• The average relative error in reconstruction was less than 2%.
• The technique effectively addresses beam-hardening artifacts.

Conclusions:

• The developed machine learning approach enables accurate quantitative CT imaging.
• Virtual monochromatic (VM) imaging has significant potential for beam-hardening correction, material differentiation, and proton therapy planning.