Full-dose whole-body PET synthesis from low-dose PET using high-efficiency denoising diffusion probabilistic model:...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Positron Emission Tomography (PET) is crucial in clinical diagnostics.
• A key challenge in PET is balancing image quality with patient radiation dose.
• Improving PET image quality while minimizing radiation exposure is essential.

Purpose of the Study:

• To introduce PET Consistency Model (PET-CM), an efficient diffusion-based method.
• To generate high-quality, full-dose PET images from low-dose PET data.
• To enhance clinical utility of PET by improving image quality and reducing radiation dose.

Main Methods:

• PET-CM utilizes a two-step diffusion process: forward diffusion with Gaussian noise and reverse diffusion using a PET Shifted-window Vision Transformer (PET-VIT) network.
• The PET-VIT network learns a consistency function for direct denoising of noise into clean, full-dose PET images.
• The method was evaluated using quantitative metrics (NMAE, PSNR, SSIM, NCC, SUV Error) and clinical assessment (Human Ranking Score).

Main Results:

• PET-CM achieved state-of-the-art image quality with significantly reduced computation time (12x faster than previous models).
• For eighth-dose to full-dose reconstruction, PET-CM yielded NMAE of 1.278%, PSNR of 33.783 dB, SSIM of 0.964, NCC of 0.968, HRS of 4.543, and SUV Error of 0.255%.
• For quarter-dose to full-dose reconstruction, results included NMAE of 0.973%, PSNR of 36.172 dB, SSIM of 0.984, NCC of 0.990, HRS of 4.428, and SUV Error of 0.151%, demonstrating high precision.

Conclusions:

• PET-CM is the first efficient diffusion-model-based method for full-dose PET image estimation from low-dose inputs.
• The method achieves comparable quality to state-of-the-art models but with superior efficiency.
• PET-CM enables high-quality, clinically relevant PET imaging while mitigating radiation risks.