Impact of Deep Learning-Based Time-of-Flight PET Images of Small Tumors Using a Human Anatomic Phantom

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Time-of-flight (ToF) in PET enhances signal-to-noise ratio, improving image quality.
• Deep learning (DL)-based ToF (DL-ToF) methods further improve tumor visibility and noise reduction in PET imaging.

Purpose of the Study:

• To quantitatively investigate the effects of DL-ToF on PET image quality using a thoracoabdominal phantom.
• To assess the impact of different DL-ToF processing levels on tumor visibility, contrast, and image characteristics.

Main Methods:

• A thoracoabdominal phantom with 18F-FDG was scanned using a BGO crystal PET/CT machine at various acquisition times.
• PET images were reconstructed using non-ToF and different levels of DL-ToF (low, middle, high).
• Quantitative analysis included SUVmean, SUVmax, and shape index maps; clinical cases were also evaluated.

Main Results:

• DL-ToF significantly improved tumor visibility and contrast, particularly with the high-precision DL (HDL) model.
• Lung tumor contrast ([Formula: see text]) increased from 3.72 (non-ToF) to 5.89 (HDL) at 10 min.
• Liver tumor contrast and delineation were also enhanced by HDL DL-ToF, with shape index maps confirming clearer boundaries.

Conclusions:

• DL-ToF processing levels influence lesion visibility and image characteristics in PET scans.
• The findings highlight the importance of understanding DL-ToF behavior for effective clinical implementation.
• High-precision DL-ToF shows promise for enhanced diagnostic accuracy in PET imaging.