ZS4D: Zero-Shot Self-Similarity-Steered Denoiser for Volumetric Photon-Counting CT

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computational Imaging

Background:

• Photon-counting computed tomography (PCCT) offers superior spatial resolution and multi-energy capabilities.
• Smaller detector elements in PCCT increase noise due to fewer detected photons.
• Effective noise reduction is crucial for high-quality PCCT image reconstruction.

Purpose of the Study:

• To introduce ZS4D, a zero-shot self-similarity-steered denoiser for PCCT.
• To improve noise suppression and structural preservation in PCCT images.
• To enhance the spatial resolution of conventional CT images using PCCT-trained models.

Main Methods:

• A self-similarity denoiser was pre-trained using self-supervised learning.
• Spectral correlations from multi-energy data and volumetric context from axial/sagittal planes were leveraged.
• The pre-trained denoiser was integrated as a prior into an iterative reconstruction framework.

Main Results:

• ZS4D effectively suppresses noise across varying noise levels in PCCT.
• Significant enhancement in image quality was observed in simulated and pre-clinical PCCT datasets.
• ZS4D demonstrated effectiveness in deblurring and improving spatial resolution of conventional CT images.

Conclusions:

• ZS4D provides a robust solution for noise reduction in PCCT reconstruction.
• The proposed method preserves structural integrity while enhancing image quality.
• Pre-training on PCCT data enables spatial resolution enhancement for conventional CT.