Diff-PCGS: physics-constrained Gaussian splatting with diffusion denoising for fast sparse-view XFCT imaging.

Area of Science:

• Medical Imaging
• Computational Physics
• Biomedical Engineering

Background:

• X-ray Fluorescence Computed Tomography (XFCT) offers element-specific, high-sensitivity imaging valuable for nanoparticle tracking and cancer diagnosis.
• Current benchtop XFCT systems face challenges with long acquisition times and image quality degradation in sparse-view scenarios.

Purpose of the Study:

• To develop a novel reconstruction framework, Diff-PCGS, for accelerating XFCT imaging while preserving high image quality.
• To address the limitations of sparse-view acquisition in XFCT by synergizing generative models and physics-based reconstruction.

Main Methods:

• Proposed Diff-PCGS framework combining a diffusion model in the projection domain for noise suppression and physics-constrained 3D Gaussian Splatting for reconstruction.
• Incorporated photon attenuation and propagation physics constraints into the iterative 3D Gaussian Splatting process.
• Validated the method using simulation data, phantom experiments, and in vivo imaging of liver tumors in mice.

Main Results:

• Diff-PCGS demonstrated superior reconstruction quality with significantly reduced projection angles (from 45 to 5).
• Outperformed conventional iterative methods in image fidelity, noise robustness, and quantitative accuracy.
• Successfully applied to in vivo sparse-view imaging of orthotopic liver tumors, confirming practical utility.

Conclusions:

• Diff-PCGS provides an effective solution for high-fidelity, fast XFCT imaging.
• Enables accurate element-specific imaging in practical scenarios, overcoming limitations of traditional methods.