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: Jul 8, 2026

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

UniOCTSeg++: Refined Hierarchical Prompt Strategy and Bi-directional Progressive Consistency Learning for Universal

Jian Zhong, Li Lin, Kenneth K Y Wong

    IEEE Transactions on Medical Imaging
    |July 6, 2026
    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

    Rationale and design of Dongzong CArdIovascuLar Bio-imaging RegistrY (DAILY) study: Bridging multiomics, imaging and cardiovascular disease.

    American journal of preventive cardiology·2026
    Same author

    ATPR promotes differentiation in gastric cancer cells by endocan.

    Molecular biology reports·2026
    Same author

    Effects of Composition and Baking Temperature on the Properties of Low-Protein Cookies.

    Food science & nutrition·2026
    Same author

    MSIDAT: an automated platform for improved metabolite annotation in mass spectrometry imaging via mass shift evaluation and customized databases.

    Analytical and bioanalytical chemistry·2026
    Same author

    Effect of starches on the preparation and properties of special low-protein scones.

    Food chemistry·2026
    Same author

    Synergistic Effects of Alkali, Salt, and Thickness Reduction on the Preparation and Properties of Low-Protein Noodles.

    Food science & nutrition·2026

    UniOCTSeg++ enhances universal medical image segmentation for OCT retinal layer analysis by improving adaptability and generalization across diverse datasets and annotation granularities. This framework achieves state-of-the-art performance and demonstrates strong label efficiency for practical deployment.

    Area of Science:

    • Medical Image Analysis
    • Computer Vision
    • Ophthalmology

    Background:

    • Universal medical image segmentation models struggle with heterogeneous datasets and varying annotation protocols.
    • Existing prompt-based methods often fail to capture background context, task dependencies, and generalize to new annotation granularities.
    • Optical Coherence Tomography (OCT) retinal layer segmentation faces challenges due to diverse annotation schemes across studies.

    Purpose of the Study:

    • To propose UniOCTSeg++, a universal OCT segmentation framework addressing limitations in current models.
    • To enhance adaptability and generalization for OCT-based retinal layer segmentation.
    • To establish a unified benchmark for evaluating universal OCT segmentation methods.

    Main Methods:

    • Introduced a Refined Hierarchical Prompting Strategy (RHPS) to create foreground-background paired embeddings, encoding anatomical relationships.

    Related Experiment Videos

    Last Updated: Jul 8, 2026

    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

  • Implemented a Bi-directional Progressive Consistency Learning (BPCL) scheme for stable predictions and reduced pseudo-label noise.
  • Constructed the Hierarchical Retinal OCT Segmentation Benchmark (HROCT-Bench) with 4.86 million OCT B-scans from eleven datasets and eight granularities.

    • Main Results:

    • UniOCTSeg++ achieved state-of-the-art adaptability, reaching 90.06% DSC and 1.38 HD95 on internal datasets.
    • External dataset performance was 86.83% DSC and 2.00 HD95.
    • With 30% labeled data and unlabeled data, UniOCTSeg++ approached fully supervised performance, showing significant label efficiency.

    Conclusions:

    • UniOCTSeg++ offers a robust and adaptable framework for universal OCT segmentation.
    • The proposed RHPS and BPCL methods effectively handle diverse annotation granularities and improve segmentation accuracy.
    • The HROCT-Bench benchmark provides a standardized evaluation for future research in universal OCT segmentation.