Related Experiment Video
Updated: Jun 21, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
21.7K
Deep Learning Approach for Differentiating Etiologies of Pediatric Retinal Hemorrhages: A Multicenter Study.
Pooya Khosravi1,2,3, Nolan A Huck1,2, Kourosh Shahraki1,2
1Department of Ophthalmology, School of Medicine, University of California, Irvine, CA 92697, USA.
International Journal of Molecular Sciences
|October 28, 2023
Summary
Deep learning models show promise in diagnosing pediatric retinal hemorrhages, differentiating between medical and trauma causes with high accuracy. AI aids diagnosis but expert ophthalmologist input remains essential.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Pediatric retinal hemorrhages present diagnostic challenges, with varied causes including trauma and medical conditions.
- Accurate etiological diagnosis is critical for patient management and legal proceedings.
- Deep learning (AI) offers potential for improved diagnostic accuracy and timeliness.
Purpose of the Study:
- To evaluate deep learning models for differentiating etiologies of pediatric retinal hemorrhages.
- To compare the performance of different AI architectures, including ResNet and transformer models.
- To assess the potential of AI in supporting ophthalmologists' diagnostic capabilities.
Main Methods:
- Analysis of 597 retinal hemorrhage fundus images from a multi-center study.
- Categorization of hemorrhages into medical and trauma etiologies.
- Application and evaluation of deep learning models like ResNet and transformer-based FastViT-SA12.
Main Results:
- FastViT-SA12 achieved 90.55% accuracy and AUC on an independent test dataset.
- ResNet18 demonstrated the highest sensitivity at 96.77%.
- The study identified areas for AI model optimization in this specific clinical context.
Conclusions:
- Deep learning models show significant potential in diagnosing pediatric retinal hemorrhages.
- AI tools can assist, but do not replace, the expertise of pediatric ophthalmologists.
- Collaboration between AI developers and clinicians is key to advancing AI in pediatric ophthalmology.
