The impact of scan time on dynamic [Formula: see text]-FAPI-04 total-body PET parametric imaging generated by deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiochemistry

Background:

• Deep learning models can generate dynamic Positron Emission Tomography (PET) parametric images from static PET scans, reducing the need for multiple scans.
• Current deep learning approaches often use static PET images from fixed time points without assessing the impact of acquisition timing.

Purpose of the Study:

• To investigate how static PET images acquired at different time points affect the quality of dynamic PET parametric images generated by deep learning.
• To determine the optimal scanning time for static PET imaging when used with deep learning.

Main Methods:

• Utilized five frames (50, 76, 80, 86, 92) from dynamic [Formula: see text]-FAPI-04 total-body PET scans as static PET inputs.
• Fed each static frame into a deep learning model to generate corresponding dynamic parametric images.
• Analyzed and compared the quality of the generated parametric images to identify optimal acquisition times.

Main Results:

• The poorest quality parametric images were generated from the 58th frame, likely due to unstable radioactive tracer diffusion.
• Frames 70-92 yielded relatively stable tracer diffusion and better quality parametric images.
• Static PET images acquired during periods of unstable tracer uptake resulted in inferior deep learning-derived dynamic PET parametric images.

Conclusions:

• The timing of static PET image acquisition significantly influences the quality of deep learning-generated dynamic PET parametric images.
• Utilizing static PET images from later scanning times, when tracer distribution is stable, is crucial for high-quality deep learning reconstructions.
• Deep learning methods for dynamic PET generation are sensitive to the tracer kinetics depicted in the input static images.