Parametric image generation with the uEXPLORER total-body PET/CT system through deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Total-body dynamic positron emission tomography/computed tomography (PET/CT) offers high sensitivity but presents challenges in generating parametric images.
• Generating parametric images typically requires lengthy dynamic scanning protocols, impacting patient comfort and throughput.

Purpose of the Study:

• To investigate the feasibility of generating parametric [Formula: see text] images directly from static PET images using a deep learning model.
• To reduce the dynamic scanning time for total-body PET/CT while maintaining image quality.
• To assess the performance of a deep learning model on a 2-m total-body PET/CT scanner (uEXPLORER).

Main Methods:

• A deep learning model was developed to predict parametric [Formula: see text] images from static [Formula: see text]F-Fluorodeoxyglucose ([Formula: see text]F-FDG) PET images.
• Image pairs were acquired from 200 patients using both a 60-min dynamic protocol and a 10-min static protocol.
• Quantitative metrics (PSNR, SSIM, NMSE) and qualitative assessment by nuclear medicine physicians were used for evaluation.

Main Results:

• The synthetic parametric PET images showed qualitative and quantitative consistency with reference images.
• The global mean structural similarity index measure (SSIM) exceeded 0.9, indicating high similarity.
• Expert physicians rated the overall subjective quality of the synthetic images highly (4.00 ± 0.45 out of 5).

Conclusions:

• The proposed deep learning technique is feasible for generating parametric images from static PET data on total-body PET/CT systems.
• This method has the potential to significantly shorten scanning durations and enhance patient experience.
• Further research is needed to validate clinical applications and understand the interpretability of the deep learning models.