Deep Few-View High-Resolution Photon-Counting CT at Halved Dose for Extremity Imaging

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• X-ray photon-counting computed tomography (PCCT) offers high-resolution, multi-energy imaging for extremities.
• Current PCCT radiation doses require optimization for improved patient safety.
• Deep learning (DL) for high-resolution volumetric PCCT reconstruction faces challenges like memory, data scarcity, and domain gaps.

Purpose of the Study:

• To develop a DL-based approach for PCCT image reconstruction with halved radiation dose and doubled speed.
• To address memory limitations, data scarcity, and domain gap issues in DL for PCCT.

Main Methods:

• A patch-based volumetric refinement network was designed to manage GPU memory constraints.
• Network training utilized synthetic data to overcome scarcity.
• Model-based iterative refinement was employed to bridge the domain gap between synthetic and clinical data.

Main Results:

• The proposed DL method enables PCCT image reconstruction at half the standard radiation dose.
• Image reconstruction speed was doubled compared to conventional methods.
• A reader study on 8 patients from a clinical trial showed no compromise in image quality or diagnostic value.

Conclusions:

• The developed DL approach shows significant potential for dose reduction in extremity PCCT.
• The method successfully addresses key challenges in applying DL to volumetric PCCT reconstruction.
• This technique offers a promising pathway for safer and faster high-resolution PCCT imaging.