Comparison of deep learning-based emission-only attenuation correction methods for positron emission tomography.

Hwang, Donghwi; Kang, Seung Kwan; Kim, Kyeong Yun; Choi, Hongyoon; Lee, Jae Sung · Eur J Nucl Med Mol Imaging · 2022

basic_science · Level V

Where this comes from

Abstract

This study aims to compare two approaches using only emission PET data and a convolution neural network (CNN) to correct the attenuation (μ) of the annihilation photons in PET. One of the approaches uses a CNN to generate μ-maps from the non-attenuation-corrected (NAC) PET images (μ-CNN<sub>NAC</sub>). In the other method, CNN is used to improve the accuracy of μ-maps generated using maximum likelihood estimation of activity and attenuation (MLAA) reconstruction (μ-CNN<sub>MLAA</sub>). We investigated the improvement in the CNN performance by combining the two methods (μ-CNN<sub>MLAA+NAC</sub>) and the suitability of μ-CNN<sub>NAC</sub> for providing the scatter distribution required for MLAA reconstruction. Image data from <sup>18</sup>F-FDG (n = 100) or <sup>68</sup> Ga-DOTATOC (n = 50) PET/CT scans were used for neural network training and testing. The error of the attenuation correction factors estimated using μ-CT and μ-CNN<sub>NAC</sub> was over 7%, but that of scatter estimates was only 2.5%, indicating the validity of the scatter estimation from μ-CNN<sub>NAC</sub>. However, CNN<sub>NAC</sub> provided less accurate bone structures in the μ-maps, while the best results in recovering the fine bone structures were obtained by applying CNN<sub>MLAA+NAC</sub>. Additionally, the μ-values in the lungs were overestimated by CNN<sub>NAC</sub>. Activity images (λ) corrected for attenuation using μ-CNN<sub>MLAA</sub> and μ-CNN<sub>MLAA+NAC</sub> were superior to those corrected using μ-CNN<sub>NAC</sub>, in terms of their similarity to λ-CT. However, the improvement in the similarity with λ-CT by combining the CNN<sub>NAC</sub> and CNN<sub>MLAA</sub> approaches was insignificant (percent error for lung cancer lesions, λ-CNN<sub>NAC</sub> = 5.45% ± 7.88%; λ-CNN<sub>MLAA</sub> = 1.21% ± 5.74%; λ-CNN<sub>MLAA+NAC</sub> = 1.91% ± 4.78%; percent error for bone cancer lesions, λ-CNN<sub>NAC</sub> = 1.37% ± 5.16%; λ-CNN<sub>MLAA</sub> = 0.23% ± 3.81%; λ-CNN<sub>MLAA+NAC</sub> = 0.05% ± 3.49%). The use of CNN<sub>NAC</sub> was feasible for scatter estimation to address the chicken-egg dilemma in MLAA reconstruction, but CNN<sub>MLAA</sub> outperformed CNN<sub>NAC.</sub>

Medical subject headings