Delta-Net: Deep Dual-Domain Alternating Optimization Network for High Pitch Helical CT Reconstruction

Area of Science:

• Medical Imaging
• Radiology
• Computer Science

Background:

• High pitch helical Computed Tomography (CT) scanning offers reduced radiation dose and improved temporal resolution.
• Incomplete scan data in high pitch CT leads to artifacts, degrading image quality and potentially impacting diagnoses.
• Current reconstruction methods have limitations in artifact suppression and image quality enhancement.

Purpose of the Study:

• To develop an advanced deep learning network for high pitch helical CT image reconstruction.
• To address the limitations of existing methods in artifact reduction and image quality.
• To improve the diagnostic accuracy of CT scans acquired with reduced radiation dose.

Main Methods:

• Proposed Delta-Net, a deep dual-domain alternating iterative optimization network.
• Introduced a novel optimization objective and an alternating iterative framework with projection and image domain corrections.
• Employed deep neural networks (IRN and PCN) for domain-specific priors and automatic hyper-parameter optimization.
• Utilized a structure-aware joint loss for artifact suppression and structure restoration.

Main Results:

• Delta-Net demonstrated superior performance in artifact suppression compared to existing methods.
• The network effectively restored fine structures in the reconstructed images.
• Evaluations on clinical datasets confirmed Delta-Net's enhanced generalization and robustness.

Conclusions:

• Delta-Net offers a significant advancement in high pitch helical CT image reconstruction.
• The dual-domain iterative approach effectively mitigates artifacts and improves image quality.
• This method holds promise for enhancing diagnostic confidence and patient safety in CT imaging.