PRIOR: Prior-Regularized Iterative Optimization Reconstruction For 4D CBCT.

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Imaging

Background:

• 4D cone-beam computed tomography (CBCT) is crucial for image-guided radiation therapy, enabling compensation for organ motion during respiration.
• Severe streaking artifacts plague 4D CBCT reconstructions due to sparse temporal projection data, compromising image quality.
• Existing reconstruction methods struggle to effectively mitigate these motion-induced artifacts.

Purpose of the Study:

• To develop an advanced reconstruction framework for 4D CBCT that addresses streaking artifacts.
• To improve the quality of 4D CBCT images for more accurate radiation therapy targeting.
• To enhance the preservation of soft tissue and fine details in reconstructed images.

Main Methods:

• Proposed a novel framework named Prior-Regularized Iterative Optimization Reconstruction (PRIOR) for 4D CBCT.
• PRIOR integrates physics-based modeling with data-driven approaches, leveraging a specialized deep learning model (PRIOR-Net).
• PRIOR-Net utilizes prior image information from fully-sampled projections to improve individual phase-resolved image reconstruction.

Main Results:

• The PRIOR framework and PRIOR-Net demonstrated significant reduction in streak artifacts in both simulated and clinical 4D CBCT datasets.
• Quantitative and qualitative evaluations showed superior performance compared to existing advanced 4D CBCT reconstruction methods.
• The proposed methods excelled in soft tissue restoration and preservation of tiny details.

Conclusions:

• The PRIOR framework and PRIOR-Net offer a robust solution for high-quality 4D CBCT reconstruction.
• These advancements hold significant potential for improving the accuracy and efficacy of image-guided radiation therapy.
• The combined physics-based and data-driven approach effectively overcomes limitations of single-model strategies.