PET image reconstruction using weighted nuclear norm maximization and deep learning prior.

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Healthcare

Background:

• Positron Emission Tomography (PET) reconstruction is inherently ill-posed, leading to limited resolution and noise.
• Deep neural networks are increasingly used to enhance PET image quality within iterative reconstruction frameworks.
• Existing methods often focus on minimizing norms, potentially overlooking image detail recovery.

Purpose of the Study:

• To propose a novel neural network-based iterative reconstruction method for PET.
• To improve image detail recovery and noise suppression in PET reconstruction.
• To introduce the application of weighted nuclear norm (WNN) maximization for PET image reconstruction.

Main Methods:

• Developed a new iterative reconstruction algorithm incorporating a neural network.
• Employed weighted nuclear norm (WNN) maximization, a novel approach in PET reconstruction.
• Utilized a neural network to manage noise arising from WNN maximization.

Main Results:

• The proposed method demonstrated superior performance on simulated and clinical PET datasets.
• Achieved an improved contrast-to-noise ratio, particularly in lesion contrast recovery.
• Successfully recovered lesions of various sizes while effectively suppressing noise in low-dose scenarios.

Conclusions:

• The novel WNN maximization approach with neural network control offers significant improvements in PET image reconstruction.
• This method provides a better trade-off between image contrast and noise reduction compared to existing techniques.
• The approach shows promise for enhanced diagnostic accuracy in low-dose PET imaging.