JNC: Deep-learning-based methods of attenuation correction for SPECT and PET (Sept/Oct 2023)-JNC: Deep-learning-based methods of attenuation correction for SPECT and PET

JNC: Deep-learning-based methods of attenuation correction for SPECT and PET

Pdf Summary

Deep learning-based methods have shown promise in the attenuation correction (AC) of SPECT and PET imaging. There are two main strategies: indirect and direct. Indirect strategies use NAC SPECT/PET images, MRI, or synthetic CT/l-maps as input to generate synthetic CT/l-maps for AC reconstruction. This approach outperforms direct strategies and can be accelerated through transfer learning. Direct strategies predict AC PET images directly from NAC PET images, with good results using CycleGAN. Joint attenuation and scatter corrections have also been addressed. Clinical evaluations show comparable or superior performance to non-deep-learning methods, but further research is needed for customization and cross-tracer and cross-scanner transfer learning. Deep learning has the potential to improve AC accuracy and efficiency in SPECT and PET imaging.

Keywords

Deep learning-based methods

attenuation correction

SPECT imaging

PET imaging

indirect strategies

direct strategies

transfer learning

CycleGAN

clinical evaluations

AC accuracy