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 Concept Videos

You might also read

Related Articles

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

Sort by
Same author

Generalized description of spectral incoherent solitons.

Optics letters·2014
Same author

Selenium-enriched Spirulina protects INS-1E pancreatic beta cells from human islet amyloid polypeptide-induced apoptosis through suppression of ROS-mediated mitochondrial dysfunction and PI3/AKT pathway.

European journal of nutrition·2014
Same author

Antitumor platinum(II) complexes of N-monoalkyl 1R,2R-diamino-cyclohexanes with 3-(nitrooxy)cyclobutane-1,1-dicarboxylate as a leaving group.

European journal of medicinal chemistry·2014
Same author

Unfolded protein response is required for the definitive endodermal specification of mouse embryonic stem cells via Smad2 and β-catenin signaling.

The Journal of biological chemistry·2014
Same author

Effects of Traditional Chinese Medicinal Plants on Anti-insulin Resistance Bioactivity of DXMS-Induced Insulin Resistant HepG2 Cells.

Natural products and bioprospecting·2014
Same author

Long noncoding RNA expression profiles of lung adenocarcinoma ascertained by microarray analysis.

PloS one·2014

Related Experiment Video

Updated: Jun 18, 2025

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

2.7K

Self-Supervised Medical Image Segmentation Using Deep Reinforced Adaptive Masking.

Zhenghua Xu, Yunxin Liu, Gang Xu

    IEEE Transactions on Medical Imaging
    |August 1, 2024
    PubMed
    Summary

    This study introduces adaptive hard masking (AHM), a novel deep reinforcement learning method to improve masked image modeling (MIM) for medical imaging. AHM enhances representation learning from unlabeled medical data, outperforming existing techniques.

    More Related Videos

    Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions
    06:18

    Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions

    Published on: April 5, 2024

    1.0K
    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
    04:48

    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

    Published on: July 5, 2024

    381

    Related Experiment Videos

    Last Updated: Jun 18, 2025

    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

    2.7K
    Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions
    06:18

    Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions

    Published on: April 5, 2024

    1.0K
    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
    04:48

    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

    Published on: July 5, 2024

    381

    Area of Science:

    • Medical Imaging
    • Computer Vision
    • Machine Learning

    Background:

    • Self-supervised learning (SSL) aims to extract transferable representations from unlabeled data.
    • Masked image modeling (MIM), inspired by NLP, shows promise in computer vision but struggles with medical images due to high redundancy and small discriminative regions.
    • Existing MIM methods often use predefined random masks, which may not be optimal for the unique characteristics of medical data.

    Purpose of the Study:

    • To propose an adaptive hard masking (AHM) approach to enhance the effectiveness of MIM in medical image analysis.
    • To leverage deep reinforcement learning to intelligently select informative regions for masking, improving representation learning.
    • To improve the performance of downstream tasks utilizing medical image representations.

    Main Methods:

    • Developed an adaptive hard masking (AHM) strategy using deep reinforcement learning.
    • Employed an asynchronous advantage actor-critic (A3C) model to predict patch-wise reconstruction loss, guiding the masking process.
    • Optimized the non-differentiable sampling of masks via reinforcement learning to focus on critical image areas.

    Main Results:

    • AHM demonstrated superior performance compared to state-of-the-art methods on two medical image datasets.
    • The approach effectively identified and utilized valuable masking regions, enhancing understanding of key areas in medical images.
    • Further experiments confirmed AHM's robustness and effectiveness in generating informative masked medical images.

    Conclusions:

    • Adaptive hard masking (AHM) significantly advances the application of masked image modeling (MIM) in the medical domain.
    • The proposed reinforcement learning-based masking strategy improves the quality of learned representations from unlabeled medical images.
    • AHM offers a promising direction for improving self-supervised representation learning in medical image analysis.