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

A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts
07:50

A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts

Published on: September 20, 2018

FGCSA-Net: A Novel Framework for Medical Report Generation Via Fine-Grained Feature Preservation and Semantic

Haiyang Zhang, Xudong Zheng, Qihan Zhang

    IEEE Journal of Biomedical and Health Informatics
    |July 6, 2026
    PubMed
    Summary

    This study introduces FGCSA-Net, a novel AI model for generating radiology reports from medical images. It enhances report accuracy by preserving visual details and improving image-text alignment.

    Related Concept Videos

    Methods of Documentation VII: EMR01:30

    Methods of Documentation VII: EMR

    Electronic Medical Records (EMRs) primarily center around electronically documenting patients' health information within a single healthcare organization or practice. They contain essential clinical data related to a patient's medical history, diagnoses, medications, treatment plans, lab results, and other pertinent information relevant to the specific encounter or episode of care. EMRs are designed to streamline documentation and workflow processes within individual healthcare settings,...

    You might also read

    Related Articles

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

    Sort by
    Same author

    Subducted sediments hidden in the upper-mantle X-discontinuities.

    Nature communications·2026
    Same author

    Trichilemmal carcinoma in China: A case report and systematic literature review.

    Rare tumors·2025
    Same author

    High-temperature structural disorders stabilize hydrous aluminosilicates in the mantle transition zone.

    Nature communications·2025
    Same author

    Identification and analysis of immune cell-related genes in cutaneous squamous cell carcinoma and drug network prediction.

    Archives of dermatological research·2024
    Same author

    Diagnosis and surgical decision-making of a 46, XX ovotesticular disorders of sex development patient: a case report.

    Frontiers in surgery·2024
    Same author

    Attention-Linear Trajectory Prediction.

    Sensors (Basel, Switzerland)·2024

    Area of Science:

    • Artificial Intelligence
    • Medical Imaging
    • Natural Language Processing

    Background:

    • Medical image report generation is a complex cross-modal task.
    • Current methods often lose subtle visual details and exhibit weak image-text alignment.
    • Accurate radiology reports are crucial for clinical decision-making.

    Purpose of the Study:

    • To develop an advanced AI model for generating high-quality radiology reports from medical images.
    • To address the limitations of existing methods in preserving visual details and aligning image features with report semantics.
    • To improve the accuracy and clinical utility of automated medical image report generation.

    Main Methods:

    • Proposed the Fine-Grained Cross-modal Semantic Alignment Network (FGCSA-Net).

    Related Experiment Videos

    Last Updated: Jul 8, 2026

    A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts
    07:50

    A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts

    Published on: September 20, 2018

  • Integrated residual feature preservation to retain diagnostically important local details.
  • Employed a cross-attention mechanism within a large-language-model framework for robust visual-text alignment.
  • Utilized residual connections for visual encoding and cross-attention for textual decoding.

    • Main Results:

    • FGCSA-Net demonstrated improved report generation quality on MIMIC-CXR and IU-Xray datasets.
    • Achieved a significant 26.7% improvement in ROUGE-L score compared to XrayGPT.
    • Effectively preserved subtle visual details and enhanced semantic alignment between images and reports.

    Conclusions:

    • FGCSA-Net offers a superior approach to medical image report generation.
    • The model's architecture effectively addresses key challenges in visual detail preservation and cross-modal alignment.
    • This advancement holds promise for improving the efficiency and accuracy of radiological diagnostics.