PET image reconstruction with deep progressive learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Convolutional neural networks (CNNs) show promise for positron emission tomography (PET) imaging but struggle with large image gaps and can degrade contrast.
• Direct learning from low-quality to high-quality PET images is challenging due to noise and contrast issues.

Purpose of the Study:

• To introduce a deep progressive learning (DPL) method for PET image reconstruction.
• To reduce background noise and enhance image contrast in PET scans.
• To overcome limitations of direct learning in PET image reconstruction.

Main Methods:

• A deep progressive learning (DPL) approach is proposed, utilizing two learning steps to bridge the gap between low and high-quality images.
• Two pre-trained neural networks are integrated into the iterative reconstruction process to manage image noise and contrast sequentially.
• A feedback structure is employed in the network design to minimize parameters, with training data from the uEXPLORER total-body PET scanner.

Main Results:

• The DPL method effectively reduces background noise in PET images.
• Contrast recovery for small lesions is significantly improved.
• Extensive phantom and patient studies confirm the efficacy of DPL for PET image quality enhancement.

Conclusions:

• Deep progressive learning (DPL) offers a promising solution for improving PET image quality by reducing noise and enhancing contrast.
• The DPL method demonstrates versatility and potential for various imaging and image processing applications.
• This approach addresses key challenges in PET reconstruction, paving the way for more accurate diagnostic imaging.