2.5D Multi-View Averaging Diffusion Model for 3D Medical Image Translation: Application to Low-Count PET...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiological Physics

Background:

• Positron Emission Tomography (PET) is crucial for clinical diagnosis but involves radiation exposure.
• Reducing tracer dose and eliminating CT scans lowers radiation but degrades PET image quality (noise, bias).
• Developing methods to translate low-dose, non-attenuation-corrected PET (NAC-LDPET) to standard-dose, attenuation-corrected PET (AC-SDPET) is essential.

Purpose of the Study:

• To develop a novel 3D image-to-image translation method for improving NAC-LDPET quality.
• To address the limitations of existing deep learning models in handling 3D PET data.
• To enable high-quality AC-SDPET reconstruction from NAC-LDPET.

Main Methods:

• Introduced a 2.5D Multi-view Averaging Diffusion Model (MADM) for 3D image translation.
• Utilized separate diffusion models for axial, coronal, and sagittal views, averaging outputs for 3D consistency.
• Incorporated a CNN-based 3D generation prior to accelerate diffusion model sampling.

Main Results:

• MADM successfully translated NAC-LDPET to AC-SDPET with high fidelity.
• The model outperformed traditional CNN-based and existing diffusion-based methods in generating 3D translation images.
• Experimental results on human patient data validated the model's effectiveness.

Conclusions:

• The novel MADM effectively generates high-quality 3D PET images from low-dose data.
• This approach offers a promising solution for reducing radiation exposure in PET imaging while maintaining diagnostic quality.
• The developed method advances deep learning applications in medical image reconstruction.