A novel supervised learning method to generate CT images for attenuation correction in delayed pet scans.

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Accurate attenuation correction is crucial for PET image reconstruction.
• Clinical PET/CT scans use CT for attenuation information, but additional scans pose cancer risks.
• Delayed PET imaging requires attenuation correction without extra CT scans.

Purpose of the Study:

• To develop a novel CT generation method for attenuation correction in delayed PET imaging.
• To eliminate the need for additional CT scans in delayed PET imaging.
• To improve the quality and quantitative accuracy of PET images.

Main Methods:

• A reconstruction network generates pseudo PET images from raw PET data.
• A second network maps PET/CT images from an initial scan to a delayed scan to generate a CT image.
• The loss function incorporates image similarity and regularization for accurate anatomy matching and deformation field smoothness.

Main Results:

• The method was evaluated using simulated and clinical PET/CT datasets.
• Generated PET images demonstrated high quality and quantitative accuracy.
• The proposed supervised learning method achieved average MAE of 4.61, RMSE of 22.75, and PSNR of 62.13 dB.

Conclusions:

• The developed method accurately generates CT images for delayed PET scan attenuation correction.
• This approach outperforms traditional methods in quantitative PET image accuracy.
• The technique reduces radiation exposure by avoiding additional CT scans.