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

MSTNet: Multi-scale spatial-aware transformer with multi-instance learning for diabetic retinopathy classification.

Xin Wei1, Yanbei Liu2, Fang Zhang2

  • 1School of Control Science and Engineering, Tiangong University, Tianjin 300387, China.

Medical Image Analysis
|February 28, 2025
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

Safety, effectiveness and treatment patterns of sodium zirconium cyclosilicate for hyperkalemia management in China: actualize study.

Frontiers in pharmacology·2026
Same author

Dapagliflozin regulates autophagy in nephropathy associated with non-diabetic obesity via elevated plasma β-hydroxybutyrate levels.

The Biochemical journal·2026
Same author

High-fidelity single-frame computational super-resolution using signal-preserving denoising-enabled deconvolution.

Nature communications·2026
Same author

The relationship between HIV infection and depression and their determinants in the MSM population in Eastern China: an analysis based on decision tree modelling and logistic regression.

BMC psychology·2026
Same author

APP ubiquitination by VHL protein is essential for MVB sorting and lysosomal degradation.

Journal of molecular cell biology·2025
Same author

Correction: High-fat diet-induced dyslipidemia drives retinal ECE-1 and ET-1 upregulation.

Frontiers in endocrinology·2025
Same journal

Generative morphodynamic forecasting enables robust zero-shot volumetric medical segmentation.

Medical image analysis·2026
Same journal

ContiMorph: An unsupervised learning framework for cardiac motion tracking with time-continuous diffeomorphism.

Medical image analysis·2026
Same journal

MedP-CLIP: Medical CLIP with region-aware prompt integration.

Medical image analysis·2026
Same journal

Multi-organ guided diagnosis of mild cognitive impairment via hierarchical alignment and knowledge distillation.

Medical image analysis·2026
Same journal

SUDA: Simultaneous unsupervised knowledge distillation and adaptation of foundation models for efficient pathological image analysis.

Medical image analysis·2026
Same journal

Beyond the LUMIR challenge: The pathway to foundational registration models.

Medical image analysis·2026
See all related articles
This summary is machine-generated.

Diabetic retinopathy (DR) detection is improved by a new AI model, MSTNet. This advanced network better analyzes subtle features in fundus images, leading to more accurate diagnoses and preventing vision loss.

Area of Science:

  • Ophthalmology
  • Artificial Intelligence
  • Medical Imaging

Background:

  • Diabetic retinopathy (DR) is a primary cause of vision loss in diabetic adults globally.
  • Early detection and treatment of DR using fundus images are crucial for preventing irreversible vision impairment.
  • Current deep learning models face challenges in capturing subtle lesion correlations and contextual information in DR fundus images due to dataset limitations.

Purpose of the Study:

  • To propose a novel Multi-scale Spatial-aware Transformer Network (MSTNet) for improved diabetic retinopathy classification.
  • To enhance the model's ability to capture both local details and global context in fundus images.
  • To address the unique characteristics of medical images, such as less distinct regions of interest, for more accurate DR assessment.

Main Methods:

Keywords:
Diabetic retinopathyMulti-instance learningMulti-scale fusionSpatial-aware moduleTransformer network

Related Experiment Videos

  • Developed MSTNet, a dual-pathway Transformer network encoding multi-scale image patches.
  • Integrated a Spatial-aware Module (SAM) to capture local spatial information.
  • Employed Multiple Instance Learning (MIL) to aggregate features and correlate with subtle lesion areas.
  • Utilized a cross-fusion classifier for final DR classification.

Main Results:

  • MSTNet demonstrated superior diagnostic and grading accuracy on four public DR datasets (APTOS2019, RFMiD2020, Messidor, IDRiD).
  • Achieved accuracy improvements of up to 2.0% (ACC) and 1.2% (F1 score) compared to existing methods.
  • Highlighted the model's effectiveness in accurately assessing fundus images for diabetic retinopathy.

Conclusions:

  • MSTNet offers a significant advancement in automated diabetic retinopathy detection and grading.
  • The proposed architecture effectively handles multi-scale features and spatial context in medical images.
  • MSTNet shows promise for clinical application in preventing vision loss due to diabetic retinopathy.