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

• Published in: European journal of nuclear medicine and molecular imaging Vol. 49; no. 6; pp. 1833 - 1842
• Main Authors: Hwang, Donghwi, Kang, Seung Kwan, Kim, Kyeong Yun, Choi, Hongyoon, Lee, Jae Sung
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.05.2022; Springer Nature B.V
• Subjects: Advanced Image Analyses (Radiomics and Artificial Intelligence); Artificial neural networks; Attenuation; Bone cancer; Cardiology; Computed tomography; Deep Learning; Emission analysis; Error correction; Fluorodeoxyglucose F18; Humans; Image processing; Image Processing, Computer-Assisted - methods; Image reconstruction; Imaging; Lesions; Lung cancer; Magnetic Resonance Imaging - methods; Maximum likelihood estimation; Medical imaging; Medicine; Medicine & Public Health; Neural networks; Nuclear Medicine; Oncology; Original Article; Orthopedics; Photons; Positron emission; Positron emission tomography; Positron Emission Tomography Computed Tomography; Positron-Emission Tomography - methods; Radiology; Scattering; Similarity; Tomography; Deep learning; Scatter correction; Simultaneous reconstruction; Attenuation correction
• ISSN: 1619-7070; 1619-7089
• DOI: 10.1007/s00259-021-05637-0

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