Poisson diffusion probabilistic model for low-dose SPECT sinogram denoising.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Processing

Background:

• Low-dose single photon emission computed tomography (SPECT) sinograms are susceptible to noise from photon attenuation, hindering accurate image reconstruction.
• Traditional denoising methods often compromise image details, which is critical for diagnostic accuracy in medical imaging.

Purpose of the Study:

• To propose a novel Poisson diffusion probabilistic model (PDPM) for denoising low-dose SPECT sinograms, addressing limitations of Gaussian noise-based models.
• To leverage the physical characteristics of Poisson noise in low-dose SPECT imaging by adapting diffusion models.

Main Methods:

• Developed a preliminary PDPM framework incorporating forward and reverse processes, adapted for Poisson noise inherent in SPECT data.
• Refined the PDPM by optimizing the training dataset generation and introducing a Temporal Prediction Aggregation Module (TPAM) to improve denoising performance.

Main Results:

• PDPM significantly improved sinogram quality, with peak signal-to-noise ratio (PSNR) increasing from 19.31 to 35.34 and structural similarity (SSIM) from 0.75 to 0.97 (p < 0.0001).
• Reconstructed images also showed substantial PSNR and SSIM improvements (p < 0.0001).
• PDPM outperformed traditional and deep learning-based denoising methods on simulated and clinical SPECT datasets, reducing coefficient of variation in regions of interest.

Conclusions:

• The proposed PDPM demonstrates effective denoising capabilities for low-dose SPECT sinograms, enhancing image quality.
• Refined PDPM framework shows promising results on both simulated and clinical data, offering a viable solution for low-dose SPECT imaging.