Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Dec 14, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

887

Full-count PET recovery from low-count image using a dilated convolutional neural network.

Karl Spuhler1, Mario Serrano-Sosa1, Renee Cattell1

  • 1Department of Biomedical Engineering, Stony Brook University, Stony Brook, NY, USA.

Medical Physics
|July 21, 2020
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

[Surgical Management of Lung Cancer in Patients with Interstitial Lung Disease].

Zhongguo fei ai za zhi = Chinese journal of lung cancer·2026
Same author

Highly Selective Methane Photooxidation to Formaldehyde by Constructing Symmetry-Breaking Sites.

Journal of the American Chemical Society·2026
Same author

Hybrid Cardiac <sup>68</sup>Ga-FAPI-4 PET/MRI in Dilated Cardiomyopathy: A Feasibility and Pilot Study.

Radiology·2026
Same author

Theta burst stimulation for suicidal ideation in depression: A systematic review and meta-analysis of efficacy and protocol moderators.

Psychiatry research·2026
Same author

Outcome comparisons of controlled ovarian hyperstimulation protocols in adenomyosis patients: a real-world retrospective cohort study.

Gynecological endocrinology : the official journal of the International Society of Gynecological Endocrinology·2026
Same author

White matter microstructural abnormalities in neurological poste-acute sequelae of coronavirus disease: Imaging signatures consistent with persistent neuroinflammation.

Brain, behavior, & immunity - health·2026

A novel dilated convolutional neural network (CNN) effectively recovers full-count Positron Emission Tomography (PET) images from low-count data. This advanced denoising method significantly improves image quality metrics compared to existing techniques.

Area of Science:

  • Medical Imaging
  • Artificial Intelligence in Healthcare
  • Nuclear Medicine

Background:

  • Positron Emission Tomography (PET) is crucial for molecular-level quantitative imaging in clinical settings.
  • Low-count PET images suffer from noise, limiting diagnostic accuracy and quantitative analysis.
  • Developing effective denoising methods is essential to enhance PET image quality and utility.

Purpose of the Study:

  • To develop a novel denoising method using a dilated convolutional neural network (CNN) for PET imaging.
  • To recover full-count PET images from low-count datasets.
  • To evaluate the performance of the proposed dilated CNN (dNet) against a standard U-Net and traditional filtering methods.

Main Methods:

  • A novel dilated CNN (dNet) was designed with hierarchical structures incorporating dilated kernels to capture larger image features.
Keywords:
U-netconvolutional neural networkdNetpositron emission tomography

More Related Videos

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

3.2K

Related Experiment Videos

Last Updated: Dec 14, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

887
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

3.2K
  • The dNet and a U-Net were trained and compared using a leave-one-out cross-validation on 35 subjects from an 18F-Fluorodeoxyglucose (FDG) PET study.
  • Low-count PET data (10% count) were generated, and image quality was assessed using Mean Absolute Percent Error (MAPE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Metric (SSIM), alongside Region of Interest (ROI) analysis.

    • Main Results:

    • Both dNet and U-Net successfully synthesized full-count PET images from low-count data, outperforming Gaussian filtering.
    • The proposed dNet demonstrated superior performance over U-Net across all objective metrics: lower MAPE (4.99 ± 0.68 vs 5.31 ± 0.76), higher PSNR (31.55 ± 1.31 dB vs 31.05 ± 1.39 dB), and higher SSIM (0.9513 ± 0.0154 vs 0.9447 ± 0.0178), with statistical significance (P < 0.01).
    • ROI quantitative analysis further confirmed the enhanced performance of dNet compared to U-Net.

    Conclusions:

    • A novel approach utilizing dilated convolutions effectively recovers high-quality full-count PET images from low-count data.
    • The developed dNet offers a significant advancement in PET image denoising, providing improved quantitative accuracy.
    • Dilated CNNs represent a promising direction for enhancing PET imaging applications.