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 Videos

Multi-Scale Dynamic Sparse Attention UNet for Medical Image Segmentation.

Xiang Li, Chong Fu, Qun Wang

    IEEE Journal of Biomedical and Health Informatics
    |March 28, 2025
    PubMed
    Summary

    This study introduces the Multi-scale Dynamic Sparse Attention (MDSA) module for medical image segmentation, improving efficiency and accuracy by focusing on relevant features. The novel MDSA-UNet model achieves competitive results without pre-training, demonstrating its effectiveness and computational efficiency.

    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

    A dual-function host-guest antioxidant system for synergistic treatment of diquat poisoning.

    RSC advances·2026
    Same author

    EECFS: Efficient Ensemble Causal Feature Selection for High-Dimensional Molecular Data.

    Journal of chemical information and modeling·2026
    Same author

    Structural network dysfunction and aberrant SC-FC coupling correlate with central post-stroke pain of thalamic origin.

    The journal of pain·2026
    Same author

    Obinutuzumab treatment for antineutrophil cytoplasmic antibody-associated vasculitis.

    Frontiers in immunology·2026
    Same author

    Contrasting greenhouse gas fluxes in a city river-lake continuum: CH<sub>4</sub> diffusive emissions offset by N<sub>2</sub>O sequestration.

    Environmental research·2026
    Same author

    AI Versus Human-Delivered Online Cognitive Behavioral Therapy for Anxiety Symptoms in Young Adults: A Randomized Controlled Trial.

    Healthcare (Basel, Switzerland)·2026

    Area of Science:

    • Medical Image Analysis
    • Deep Learning
    • Computer Vision

    Background:

    • Transformers excel at long-range dependencies in medical imaging.
    • Background noise and computational burden challenge transformer-based segmentation.
    • Preserving fine-grained details is crucial for complex medical images.

    Purpose of the Study:

    • To develop a novel attention module for efficient and accurate medical image segmentation.
    • To address the computational challenges posed by background noise in transformer models.
    • To maintain fine-grained detail preservation in segmentation tasks.

    Main Methods:

    • Proposed the Multi-scale Dynamic Sparse Attention (MDSA) module.
    • Incorporated multi-scale aggregation and coarse-grained filtering before fine-grained self-attention.

    Related Experiment Videos

  • Developed MDSA-UNet using MDSA, enhanced downsampling merging (EDM), and enhanced upsampling fusion (EUF) modules.

    • Main Results:

    • MDSA-UNet achieved high segmentation performance across four datasets (DDTI, TN3K, ISIC2018, ACDC) without pre-training.
    • Achieved Dice scores of 82.10% (DDTI), 80.20% (TN3K), 90.75% (ISIC2018), and 91.05% (ACDC).
    • Maintained computational efficiency with 6.65M parameters and 4.54G FLOPs at 224x224 resolution.

    Conclusions:

    • MDSA-UNet offers a computationally efficient and effective solution for medical image segmentation.
    • The dynamic sparse attention mechanism successfully reduces computational load while preserving critical details.
    • The model demonstrates strong performance, competing with pre-trained methods without requiring pre-training.