摘要
Attenuation correction can improve the contrast of the positron emission tomography (PET) images and provide valuable information for the diagnosis. Clinically, attenuation correction is performed with computed tomography (CT) or magnetic resonance (MR) images, which requires an additional CT or MR scan. In order to save the cost of scanning and reduce the harm of CT radiation to patients, we proposed invertible AC-flow to obtain attenuation corrected (AC) PET images directly from non-attenuation corrected (NAC) PET images. The overall AC-flow adopted multi-slice inputs to utilize the spatial correlation of adjacent slices, and was consisted of 8 invertible blocks, and each invertible block was made up of a invertible 1 x 1 convolution layer and a enhanced affine coupling layer. Several small U-nets were adopted to learn the transformation parameters of the enhanced affine coupling layers. The proposed AC-flow was validated on a cardiac gated-PET dataset of N-13-Ammonia, and the experimental results showed that the AC-flow was able to obtain AC PET images of high quality, and achieved higher SSIM (0:9202 +/- 0:0592) and PSNR (27:60 +/- 5:10) than DCNN (SSIM: 0:8957 +/- 0:0814, PSNR: 26:66 +/- 6:41) and DeepDAC (SSIM: 0:8786 +/- 0:1027, PSNR: 25:34 +/- 6:35).
学科与关键词
- Deep-learning-based methods of attenuation correction for SPECT and PETChen, Xiongchao; Liu, Chi*Journal of Nuclear Cardiology, 2023, 30(5): 1859-1878.被引 13
- DeTransUnet: attenuation correction of gated cardiac images without structural informationWang, Bo; Lu, Lijun*; Liu, Huafeng*Physics in Medicine and Biology, 2022, 67(16): 165007.被引 2
- Transfer learning-based attenuation correction for static and dynamic cardiac PET using a generative adversarial networkSun, Hao; Wang, Fanghu; Yang, Yuling; Hong, Xiaotong; Xu, Weiping; Wang, Shuxia*; Mok, Greta S. P.*; Lu, Lijun*European Journal of Nuclear Medicine and Molecular Imaging, 2023, 50(12): 3630-3646.被引 1
