Deep learning-based techniques for estimating high-quality full-dose positron emission tomography images from...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose Positron Emission Tomography (PET) imaging is desirable to reduce radiation exposure.
• Synthesizing high-quality PET images from low-dose data remains a challenge.

Purpose of the Study:

• To systematically review the potential of deep learning algorithms for converting low-dose PET images to full-dose quality.
• To evaluate the effectiveness of various deep learning models across different body regions.

Main Methods:

• Systematic literature search of PubMed, Web of Science, Scopus, and IEEE databases (2017-2023).
• Inclusion of 55 articles utilizing deep learning models like Generative Adversarial Networks (GANs) and UNET.
• Analysis of diverse datasets, preprocessing techniques, input data, and loss functions.

Main Results:

• Deep learning models demonstrated significant potential in synthesizing high-quality PET images from low-dose inputs.
• Quantitative and qualitative evaluations, including physician assessments, supported the efficacy of these methods.
• The reviewed studies indicate successful application across various body regions.

Conclusions:

• Deep learning algorithms offer a promising approach for generating high-quality PET images from low-dose acquisitions.
• This technology has the potential to enhance clinical utility and patient safety in PET imaging.