Generation of Structural MR Images from Amyloid PET: Application to MR-Less Quantification - PubMed Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2018 Jul;59(7):1111-1117.
doi: 10.2967/jnumed.117.199414. Epub 2017 Dec 7.

Generation of Structural MR Images from Amyloid PET: Application to MR-Less Quantification

Hongyoon Choi  1 , Dong Soo Lee; Alzheimer’s Disease Neuroimaging Initiative
Affiliations

Generation of Structural MR Images from Amyloid PET: Application to MR-Less Quantification

Hongyoon Choi et al. J Nucl Med. 2018 Jul.

Abstract

Structural MR images concomitantly acquired with PET images can provide crucial anatomic information for precise quantitative analysis. However, in the clinical setting, not all the subjects have corresponding MR images. Here, we developed a model to generate structural MR images from amyloid PET using deep generative networks. We applied our model to quantification of cortical amyloid load without structural MR. Methods: We used florbetapir PET and structural MR data from the Alzheimer Disease Neuroimaging Initiative database. The generative network was trained to generate realistic structural MR images from florbetapir PET images. After the training, the model was applied to the quantification of cortical amyloid load. PET images were spatially normalized to the template space using the generated MR, and then SUV ratio (SUVR) of the target regions was measured by predefined regions of interest. A real MR-based quantification was used as the gold standard to measure the accuracy of our approach. Other MR-less methods-a normal PET template-based, a multiatlas PET template-based, and a PET segmentation-based normalization/quantification-were also tested. We compared the performance of quantification methods using generated MR with that of MR-based and MR-less quantification methods. Results: Generated MR images from florbetapir PET showed signal patterns that were visually similar to the real MR. The structural similarity index between real and generated MR was 0.91 ± 0.04. The mean absolute error of SUVR of cortical composite regions estimated by the generated MR-based method was 0.04 ± 0.03, which was significantly smaller than other MR-less methods (0.29 ± 0.12 for the normal PET template, 0.12 ± 0.07 for the multiatlas PET template, and 0.08 ± 0.06 for the PET segmentation-based methods). Bland-Altman plots revealed that the generated MR-based SUVR quantification was the closest to the SUVRs estimated by the real MR-based method. Conclusion: Structural MR images were successfully generated from amyloid PET images using deep generative networks. Generated MR images could be used as templates for accurate and precise amyloid quantification. This generative method might be used to generate multimodal images of various organs for further quantitative analyses.

Keywords: MR generation; PET quantification; deep learning; florbetapir PET; generative adversarial network.

PubMed Disclaimer

Figures

FIGURE 1.
FIGURE 1.
Adversarial training for MR generation network. GAN consists of multiple convolutional and deconvolutional layers to translate florbetapir PET to structural MR images. Training of network was aimed at generating MR images, which cannot be distinguished from real images. In contrast, another discriminator network was trained to distinguish real MR from generated MR images. They competed in entire training process.
FIGURE 2.
FIGURE 2.
Amyloid PET quantification using different methods. We applied MR generation model to quantification of amyloid PET. As gold standard method, MR-based normalization was used. PET images were coregistered to corresponding MR and then nonrigid transformation of MR was performed for spatial normalization. Predefined cortical and reference regions were used for calculating SUVR. For normal PET template–based method, averaged florbetapir PET images of normal controls were used as template, and then all PET images were directly normalized to this template. Multiatlas PET template–based quantification chose PET template most similar to a subject’s PET image among various PET templates with different tracer uptake patterns, and then images were normalized to selected templates. In addition, as modified method, PET was directly used for tissue segmentation and segmented tissues were normalized into template space. As application of our GAN model, generated MR images were spatially normalized to MR template, and corresponding PET images were transformed to template space. We compared SUVRs measured by these 4 different normalization methods.
FIGURE 3.
FIGURE 3.
Examples of generated MR images. After training, MR images were generated from amyloid PET images of independent test set. Regardless of subjects’ diagnosis, MR images were generated, and signal patterns similar to corresponding real MR images were observed. Quantitative similarity measured by structural similarity index measurement between real and generated brain was 0.91 ± 0.04.
FIGURE 4.
FIGURE 4.
Scatterplots of SUVRs calculated by different normalization methods. SUVRs measured by MR-less methods were compared with MR-based quantification results. Generated MR-based SUVR quantification results were highly correlated with MR-based quantification results. However, normal PET template–based method showed biased results. Multiatlas PET template–based and PET segmentation–based methods showed less biased results than normal PET template–based method, however, relatively higher error than generated MR-based method. CN = controls.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Werner P, Barthel H, Drzezga A, Sabri O. Current status and future role of brain PET/MRI in clinical and research settings. Eur J Nucl Med Mol Imaging. 2015;42:512–516. - PubMed
    1. Pichler BJ, Kolb A, Nägele T, Schlemmer H-P. PET/MRI: paving the way for the next generation of clinical multimodality imaging applications. J Nucl Med. 2010;51:333–336. - PubMed
    1. Clark CM, Schneider JA, Bedell BJ, et al. Use of florbetapir-PET for imaging β-amyloid pathology. JAMA. 2011;305:275–283. - PMC - PubMed
    1. Johnson KA, Minoshima S, Bohnen NI, et al. Appropriate use criteria for amyloid PET: a report of the Amyloid Imaging Task Force, the Society of Nuclear Medicine and Molecular Imaging, and the Alzheimer’s Association. J Nucl Med. 2013;54:476–490. - PubMed
    1. Lopresti BJ, Klunk WE, Mathis CA, et al. Simplified quantification of Pittsburgh compound B amyloid imaging PET studies: a comparative analysis. J Nucl Med. 2005;46:1959–1972. - PubMed

Publication types