DECT sparse reconstruction based on hybrid spectrum data generative diffusion model

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Dual-energy computed tomography (DECT) offers material differentiation but faces challenges with radiation exposure.
• Sparse view DECT imaging reduces radiation dose but can compromise image quality.
• Existing reconstruction methods struggle to balance image quality and radiation dose in DECT.

Purpose of the Study:

• To develop a novel reconstruction model for sparse view DECT imaging.
• To improve image quality in DECT while minimizing radiation exposure.
• To address the trade-off between image quality and radiation dose in DECT.

Main Methods:

• A hybrid spectrum data generative diffusion reconstruction model (HSGDM) was developed.
• The model leverages spectral similarity using interleaved angles for sparse scanning.
• It employs a hybrid constraint integrating image and wavelet space diffusion models for iterative reconstruction.

Main Results:

• The HSGDM achieved competitive precision in CT values, detail preservation, and artifact elimination.
• Reconstruction with 30 sparse views showed significant improvements in PSNR, SSIM, and FID scores.
• Ablation studies confirmed the effectiveness of the hybrid prior combining image and wavelet space modules.

Conclusions:

• A unified, optimized mathematical model integrating image and wavelet space priors was developed.
• The proposed HSGDM offers a practical and interpretable solution for sparse DECT reconstruction.
• Experimental results validate the model's superior performance in sparse DECT imaging.