Positron emission tomography (PET) imaging is a noninvasive technique that makes it possible to probe biological metabolic processes in vivo. It detects gamma photons emitted by radionuclides in the organs or tissues of interest and then visually represents their metabolic activity.  

However, effectively and accurately reconstructing PET images is challenging due to the ill-posedness of the inverse problem and various physical degradation factors. Many image reconstruction methods have been proposed over the past few years to improve diagnostic performance. However, most of these methods can compromise the reconstruction of important high-frequency structural details after aggressive denoising.  

Recently, deep learning has been successfully applied to many areas of medical imaging, and achieved better performance than the previous state-of-the-art methods. 

Researchers from the Shenzhen Institutes of Advanced Technology (SIAT) of the Chinese Academy of Sciences developed a novel deep learning method that reconstructed PET images using a cascading back-projection neural network (bpNet). The study was published in IEEE Journal of Selected Topics in Signal Processing on May 29. 

This network consisted of a domain translation operation, which actes as prior knowledge, cascaded with a modified encoder-decoder network.  

Based on the embedded domain transformation operations, when the imaging geometry changed, the network simply needed to input new geometric parameters, no retraining was required. The image reconstruction pipeline ranged from the sinogram to the back-projection image and then to the PET image.  

The results demonstrated that bpNet provided favorable reconstructed image quality, especially for low-count PET image reconstruction and for the next step, the team will evaluate the application of this method in clinical PET data. 

Schematic of the bpNet architecture. (Image by SIAT) 

Comparison of reconstruction results. (Image by SIAT)

Media Contact:
ZHANG Xiaomin
Email: xm.zhang@siat.ac.cn