Related Experiment Video
Updated: Jun 22, 2026
06:27
Transcutaneous Microcirculatory Imaging in Preterm Neonates
Published on: December 31, 2015
8.3K
DeepQuality improves infant retinopathy screening
Longhui Li1, Duoru Lin2, Zhenzhe Lin1
1State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangdong Provincial Key Laboratory of Ophthalmology and Visual Science, Guangdong Provincial Clinical Research Center for Ocular Diseases, Guangzhou, China.
NPJ Digital Medicine
|October 16, 2023
Summary
A new AI system, DeepQuality, assesses and enhances infantile fundus images, improving infant retinopathy screening. This technology addresses image quality issues, boosting diagnostic accuracy for clinicians and AI models.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Image quality significantly impacts AI diagnostic model performance in clinical settings.
- Infantile fundus photography often suffers from poor image quality due to patient cooperation, increasing misdiagnosis risk.
- Retinopathy of prematurity (ROP) screening requires high-quality fundus images for accurate diagnosis.
Purpose of the Study:
- To develop a deep learning-based system (DeepQuality) for assessing and enhancing infantile fundus image quality.
- To improve the accuracy of infant retinopathy screening through enhanced image quality.
- To evaluate the impact of DeepQuality on clinical diagnosis and AI model performance.
Main Methods:
- Development of a deep learning model for detecting quality defects (integrity, illumination, clarity) in infantile fundus images.
- Calculation of Area Under the Curve (AUC) values to assess the accuracy of quality defect detection.
- Analysis of a large dataset (2,015,758 images) to determine the prevalence of image quality defects.
- Implementation of a quality enhancement module within DeepQuality.
- Evaluation of the impact of image enhancement on clinician and AI model performance for ROP diagnosis.
Main Results:
- DeepQuality achieved high accuracy in detecting quality defects, with AUC values ranging from 0.933 to 0.995.
- 58.3% of analyzed infantile fundus images exhibited quality defects, with significant variations across hospitals.
- Quality enhancement by DeepQuality significantly improved clinician performance in diagnosing ROP.
- Integrating DeepQuality with AI diagnostic models enhanced ROP detection capabilities.
Conclusions:
- DeepQuality is an effective tool for assessing and enhancing infantile fundus image quality, crucial for accurate ROP screening.
- The prevalence of image quality issues in real-world infantile fundus photography is substantial.
- DeepQuality improves both human and AI diagnostic performance, offering a valuable reference for future image-based screening systems.
