End-to-End Differentiable Photon Counting CT

Area of Science:

• Medical Imaging
• Computational Imaging
• Photon-Counting CT

Background:

• Quantitative imaging is crucial for spectral X-ray and CT systems.
• Photon-counting CT (PCCT) achieves quantitative imaging via material decomposition (MD).
• Current methods often require manual intervention and lack end-to-end optimization.

Purpose of the Study:

• To present a novel framework for end-to-end differentiable PCCT imaging.
• To enable cross-domain learning and optimization for upstream models using quantitative image data.
• To automate and adapt solutions for imaging tasks, achieving quantitative imaging computationally.

Main Methods:

• Developed a differentiable framework for the PCCT imaging chain.
• Made material decomposition (MD) differentiable using the Implicit Function Theorem.
• Integrated the differentiable MD as a layer for end-to-end optimization.

Main Results:

• The framework allows end-to-end training without direct-domain supervision.
• Demonstrated applicability in correcting detector energy bin drift.
• Successfully trained an object scatter correction network using quantitative material images.

Conclusions:

• The differentiable PCCT framework facilitates quantitative imaging through computation.
• Enables adaptive solutions for diverse imaging tasks by leveraging quantitative image information.
• Offers a powerful approach for optimizing imaging systems and improving image quality.