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: Aug 21, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

1.9K

Benchmarking and Boosting Transformers for Medical Image Classification.

DongAo Ma1, Mohammad Reza Hosseinzadeh Taher1, Jiaxuan Pang1

  • 1Arizona State University, Tempe, AZ 85281, USA.

Domain Adaptation and Representation Transfer : 4Th MICCAI Workshop, DART 2022, Held in Conjunction with MICCAI 2022, Singapore, September 22, 2022, Proceedings. Domain Adaptation and Representation Transfer (Workshop) (4Th : 2022 : Sin
|November 16, 2022
PubMed
Summary
This summary is machine-generated.

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

Decrypting the allosteric network: multi-scale dynamics of SARS-CoV-2 main protease inhibition by <i>Moringa oleifera</i> flavonoids.

Journal of biomolecular structure & dynamics·2026
Same author

Cardiac Myeloid Sarcoma Arising From a Mediastinal Mass Encasing the Aorta: Diagnostic Pitfalls and Challenges.

JACC. Case reports·2026
Same author

Diffuse Bilateral Pulmonary Nodules and Calcifications during Third Kidney Transplant Evaluation.

Mayo Clinic proceedings·2026
Same author

Interlobular Septal Thickening: A Pictorial Review.

Journal of computer assisted tomography·2026
Same author

Autodidactic dense anatomical models.

Medical image analysis·2026
Same author

Ictal-interictal SPECT analysis using SPM and multimodal prediction of epilepsy surgery outcome.

Journal of clinical neuroscience : official journal of the Neurosurgical Society of Australasia·2026
Same journal

Discriminative, Restorative, and Adversarial Learning: Stepwise Incremental Pretraining.

Domain adaptation and representation transfer : 4th MICCAI Workshop, DART 2022, held in conjunction with MICCAI 2022, Singapore, September 22, 2022, proceedings. Domain Adaptation and Representation Transfer (Workshop) (4th : 2022 : Sin...·2022
Same journal

POPAR: Patch Order Prediction and Appearance Recovery for Self-supervised Medical Image Analysis.

Domain adaptation and representation transfer : 4th MICCAI Workshop, DART 2022, held in conjunction with MICCAI 2022, Singapore, September 22, 2022, proceedings. Domain Adaptation and Representation Transfer (Workshop) (4th : 2022 : Sin...·2022
See all related articles

Visual transformers show promise in medical imaging, outperforming convolutional neural networks (CNNs) when properly initialized. Self-supervised learning and domain-specific data are key for transformer success in this field.

Area of Science:

  • Computer Vision
  • Medical Imaging Analysis
  • Deep Learning

Background:

  • Visual transformers are increasingly outperforming convolutional neural networks (CNNs) in general computer vision benchmarks.
  • The application and performance of visual transformers in the specialized domain of medical imaging remain under-explored.
  • Challenges in medical imaging include data scarcity and the domain gap between natural and medical images.

Purpose of the Study:

  • To benchmark visual transformers against CNNs for medical image classification.
  • To investigate the impact of different pre-training strategies (supervised and self-supervised) on transformer performance in medical imaging.
  • To propose and evaluate a method for bridging the domain gap using unlabeled in-domain data.

Main Methods:

Keywords:
BenchmarkingDomain-adaptive pre-trainingTransfer learningVision Transformer

More Related Videos

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

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

481

Related Experiment Videos

Last Updated: Aug 21, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

1.9K
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

481
  • Benchmarking various visual transformer architectures with different pre-training methods against established CNNs.
  • Utilizing self-supervised learning, specifically masked image modeling, for pre-training.
  • Employing unlabeled, large-scale in-domain medical data for continuous pre-training to bridge the domain gap.

    • Main Results:

    • Initialization quality is more critical for transformers than for CNNs in medical imaging tasks.
    • Self-supervised learning, particularly masked image modeling, yields more generalizable representations compared to supervised pre-training.
    • Leveraging larger, domain-specific datasets through self-supervised continuous pre-training effectively reduces the domain gap.

    Conclusions:

    • Visual transformers offer competitive or superior performance in medical image classification, contingent on effective pre-training and initialization.
    • Self-supervised learning strategies are crucial for maximizing the potential of transformers in data-limited medical domains.
    • Domain adaptation techniques, such as self-supervised continuous pre-training on in-domain data, are vital for successful transformer deployment in medical imaging.