We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Physics > Medical Physics

Title: Anatomical-Guided Attention Enhances Unsupervised PET Image Denoising Performance

Abstract: Although supervised convolutional neural networks (CNNs) often outperform conventional alternatives for denoising positron emission tomography (PET) images, they require many low- and high-quality reference PET image pairs. Herein, we propose an unsupervised 3D PET image denoising method based on anatomical information-guided attention mechanism. Our proposed magnetic resonance-guided deep decoder (MR-GDD) utilizes the spatial details and semantic features of MR-guidance image more effectively by introducing encoder-decoder and deep decoder subnetworks. Moreover, the specific shapes and patterns of the guidance image do not affect the denoised PET image, because the guidance image is input to the network through an attention gate. Monte Carlo simulation using the [$^{18}$F]fluoro-2-deoxy-D-glucose (FDG) shows that the proposed method outperforms other denoising algorithms in terms of the highest peak signal-to-noise ratio and structural similarity (28.33 dB/0.886). Furthermore, we experimentally visualized the behavior of the optimization process, which is often unknown in unsupervised CNN-based restoration problems. For preclinical (using [$^{18}$F]FDG and [$^{11}$C]raclopride) and clinical (using [$^{18}$F]florbetapir) studies, the proposed method demonstrates state-of-the-art denoising performance while retaining spatial resolution and quantitative accuracy, despite using only a single architecture for various noisy PET images with 1/10th of the full counts. These results suggest that the proposed MR-GDD can reduce PET scan times and PET tracer doses considerably without impacting patients.
Comments: 24 pages, 11 figures
Subjects: Medical Physics (physics.med-ph); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Journal reference: Med. Image Anal. (2021)
DOI: 10.1016/j.media.2021.102226
Cite as: arXiv:2109.00802 [physics.med-ph]
  (or arXiv:2109.00802v1 [physics.med-ph] for this version)

Submission history

From: Fumio Hashimoto [view email]
[v1] Thu, 2 Sep 2021 09:27:07 GMT (2100kb)
[v2] Wed, 8 Sep 2021 01:53:01 GMT (2311kb)

Link back to: arXiv, form interface, contact.