A tissue-informed deep learning-based method for positron range correction in preclinical [Formula: see text]Ga PET...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Nuclear Medicine

Background:

• Positron range (PR) blurs PET images and reduces accuracy, especially with high-energy isotopes like [Formula: see text]Ga.
• Accurate quantitative PET imaging is crucial for diagnosis and treatment monitoring.

Purpose of the Study:

• To develop a deep learning approach for positron range correction (PRC) in PET imaging.
• To improve spatial resolution and quantitative accuracy using 3D residual encoder-decoder convolutional neural networks (3D RED-CNNs).

Main Methods:

• Three 3D RED-CNN architectures (Single-Channel, Two-Channel, DualEncoder) were trained on simulated PET data.
• Models were evaluated using simulated and real [Formula: see text]Ga PET data, comparing against Richardson-Lucy deconvolution (RL-PRC).
• Performance metrics included MAE, SSIM, CR, and CNR.

Main Results:

• CNN methods improved SSIM by 19% and reduced MAE by 13% compared to RL-PRC in simulations.
• The Two-Channel model achieved superior contrast recovery (97% in lung) and contrast-to-noise ratio.
• CNN models maintained stable noise levels, unlike RL-PRC which increased noise.

Conclusions:

• CNN-based PRC, particularly the Two-Channel model, significantly enhances quantitative PET imaging for [Formula: see text]Ga.
• This deep learning approach surpasses conventional deconvolution methods in both simulated and real preclinical data.
• Future work aims to improve model generalization and apply it to other high-energy PET isotopes.