Small animal PET to CT image synthesis based on conditional generation network

During the image reconstruction process of Positron Emission Computed Tomography (PET), the tissue of the object being photographed will undergo Compton scattering with photons generated by the annihilation of the positive and negative electrons, which will cause the reconstructed image to be blurred. Traditional PET equipment will add a Computed Tomography (CT) equipment module. During PET shooting process, CT images are taken at the same time, and the CT images are used for attenuation correction to reduce scattering effect of PET. Especially in PET equipment that requires high accuracy, attenuation correction has a greater impact on the quality of the final PET reconstruction. The CT image is generally used for attenuation correction, not for final diagnosis. The additional CT shooting will bring more radiation to the shooting target and increase manufacturing cost of the equipment. This paper proposes a method of generating Synthetic CT image from PET based on Generative Adversarial Network (GAN). The generator of our network uses an improved generator based on encoder-decoder structure, and the discriminator has a local field of view. This article compares the effects of several loss functions respectively. These loss functions successfully improve the model's capabilities and results. The quantitative and qualitative results show that our proposed method outperforms common methods.