A CT-free deep-learning-based attenuation and scatter correction for copper-64 PET in different time-point scans

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Whole-body 64Cu PET imaging requires accurate attenuation and scatter correction for reliable diagnostics.
• Traditional correction methods can be time-consuming or require additional hardware.
• Deep learning offers a promising avenue for automated and efficient image correction.

Purpose of the Study:

• To develop and evaluate a deep learning model for attenuation and scatter correction (PET-DLAC) in whole-body 64Cu PET imaging.
• To assess the model's performance using transfer learning from a pre-trained model on 68Ga-PSMA PET images.
• To compare the quality of deep learning-corrected PET images with conventional CT-based attenuation correction (PET-CTAC).

Main Methods:

• Implementation of a swinUNETR deep learning model using the MONAI framework.
• Training the model on whole-body PET-nonAC and PET-CTAC image pairs.
• Fine-tuning a model pre-trained on 68Ga-PSMA PET images using a limited dataset of 15 64Cu PET images.
• Evaluation of the model on six independent 64Cu PET datasets at different time points (1h, 12h, 48h).

Main Results:

• The deep learning model achieved high image quality, with excellent performance at the 12-hour time point (MSE: 0.002 ± 0.0004 SUV², PSNR: 43.14 ± 0.08 dB, SSIM: 0.981 ± 0.002).
• Consistent and strong results were observed across different time points (1h, 12h, 48h), indicating robustness.
• The model successfully generated PET-DLAC images closely resembling PET-CTAC images, demonstrating minimal errors despite a small training dataset.

Conclusions:

• The proposed deep learning model effectively performs attenuation and scatter correction in whole-body 64Cu PET imaging.
• Transfer learning is a viable strategy to achieve acceptable performance even with limited specific training data.
• This AI-driven approach holds potential for improving the efficiency and accuracy of 64Cu PET image analysis.