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

An Adaptive Deep Learning Framework for Multi-Label Chest X-Ray Diagnosis Using a Hybrid CNN-Transformer Architecture

Chi-Feng Hsieh1,2, Hsu-Hsia Peng1, Yu-Hsiang Tsai2

  • 1Department of Biomedical Engineering and Environmental Sciences, National Tsing Hua University, Hsinchu 300, Taiwan.

Diagnostics (Basel, Switzerland)
|May 4, 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

Strontium-crosslinked potassium-alginate beads as coordination-driven ion-exchange biomaterials for regenerative microenvironment engineering.

International journal of biological macromolecules·2026
Same author

Quantitative Assessment of Myocardial Velocity and Dyssynchrony in Fontan Circulation Using MR Tissue Phase Mapping.

NMR in biomedicine·2026
Same author

Cross-Dataset Generalization of Deep Learning-Based Detectors for Intracranial Hemorrhage Subtype Localization on Noncontrast Head CT: A Comparative Study.

Diagnostics (Basel, Switzerland)·2026
Same author

Nervous necrosis virus B2 non-structural protein shut off host translation via sequestration of polyadenylate-binding protein.

Biochemical and biophysical research communications·2026
Same author

Advances in nanotechnology for addressing biofilm-related challenges in medical devices.

Nanotechnology·2026
Same author

Reframing arterial transit artefacts as signals of vascular diversity.

European radiology·2026

A novel deep learning framework using hybrid CNN-transformer models and ensemble strategies showed modest but significant improvements in classifying thoracic diseases on chest X-rays. External validation confirmed partial generalizability, highlighting the potential of these advanced AI techniques in medical imaging.

Area of Science:

  • Artificial Intelligence
  • Medical Imaging
  • Radiology

Background:

  • Deep learning models are crucial for analyzing medical images.
  • Multi-label classification of thoracic diseases on chest radiographs presents significant challenges.
  • Developing robust and generalizable AI frameworks is essential for clinical applications.

Purpose of the Study:

  • To develop and externally validate a deep learning framework for multi-label thoracic disease classification on chest radiographs.
  • To leverage hybrid convolutional neural network (CNN)-transformer architectures, hierarchical scalar-weighted fusion, and ensemble strategies.
  • To benchmark performance against established models and assess generalizability across diverse datasets.

Main Methods:

  • A retrospective, multi-center study using NIH ChestX-ray14, CheXpert, and ChestX-Det10 datasets.
Keywords:
chest radiographydeep learninghybrid CNN–transformermulti-label classificationthoracic disease

Related Experiment Videos

  • Development of nine CNN-transformer hybrid models incorporating multi-scale DenseNet121 features, scalar-weighted fusion, and cross-attention.
  • Systematic benchmarking and evaluation using AUROC, precision, recall, F1-score, and other metrics.
  • Application of post hoc ensemble methods, including class-wise Top-3 Grid Search, for performance enhancement.

    • Main Results:

    • The proposed hybrid model achieved a mean AUROC of 0.8495 on the internal test set, outperforming the DenseNet121 baseline.
    • The Top-3 Grid Search ensemble further improved internal performance to a mean AUROC of 0.8577.
    • External validation demonstrated consistent outperformance over DenseNet121, with mean AUROCs of 0.6500 (CheXpert) and 0.6592 (ChestX-Det10).
    • Significant improvements were observed for specific conditions like cardiomegaly, mass, and pneumothorax, with Grad-CAM confirming alignment with relevant regions.

    Conclusions:

    • The CNN-transformer framework with ensemble strategies offers modest but statistically significant improvements in multi-label chest X-ray classification.
    • External validation indicates partial generalizability, though performance is moderate under domain shifts.
    • The study underscores the potential of advanced AI architectures for enhancing diagnostic accuracy in thoracic radiology.