Loading...

Scattering Correction in PET Device using Deep Learning

Kamalizadeh, Zeynab | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57517 (46)
  4. University: Sharif University of Technology
  5. Department: Energy Engineering
  6. Advisor(s): Vosoghi, Naser; Ghaffarian, Pardis; Sheikhzadeh, Peyman
  7. Abstract:
  8. Positron Emission Tomography (PET) imaging is a modern sectional imaging technique widely used due to its advantages in visualizing the physiological functions of body tissues. In this method, a radiotracer is injected into the body, concentrating in the area of interest. As the positron-emitting radionuclide decays, positron annihilation occurs, resulting in the emission of two gamma rays in opposite directions. The PET scanner detects these gamma rays; however, scattering may occur during their journey from the patient’s body to the detector, which can introduce errors in tumor localization. To reconstruct images and correct for these scatter effects, various techniques such as iterative reconstruction, Gaussian filtering, optimization of the absorption process, statistical methods, and regularization are employed. These methods, however, tend to be slow and involve complex computations. In contrast, deep learning methods offer a faster and more accurate alternative. Deep learning approaches have proven to be effective in addressing attenuation and scatter correction in PET studies, outperforming conventional techniques and enhancing the quality of PET images. We explore various neural network architectures, including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Deep Belief Networks (DBN). The CNN framework takes the PET images as input, estimates the factors of attenuation and scattering, and performs corrections on the images. Different configurations of CNN have been utilized for scatter correction and image reconstruction in PET systems. CNNs have shown promise in improving the accuracy and efficiency of scatter correction and attenuation tasks. In this study, leveraging the power of CNNs and employing algorithms such as U-Net, Res-Net, and Seg-Net, scatter correction was carried out on brain images obtained from the PET imaging system. Ultimately, based on metrics such as PSNR, SSIM, and RMSE derived from each algorithm, it was concluded that the U-Net algorithm, with a linear activation function in its last layer, demonstrated superior performance in scatter correction compared to the other algorithms
  9. Keywords:
  10. Positron Emission Tomography (PET) ; Scatter Correction ; U-Net Model ; Seg-Net Algorithm ; Res-Net Algorithm ; Convolutional Neural Network ; Recurrent Neural Networks

 Digital Object List

 Bookmark

No TOC