Related Experiment Video
Updated: May 1, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Image-Based Deep Learning for Cataract Diagnosis: Systematic Review and Meta-Analysis
Ruixi Li1, Hongyi Li1, Chong Li2
1Department of Ophthalmology, The Second Affiliated Hospital of Kunming Medical University, Kunming, Yunnan Province, China.
Journal of Medical Internet Research
|April 29, 2026
Summary
Deep learning (DL) shows high accuracy in detecting and classifying cataracts from images, comparable to human experts. However, limited external validation necessitates caution for widespread clinical use.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Cataracts are a leading cause of blindness globally, necessitating early diagnosis.
- Effective diagnostic tools are crucial for managing this high-prevalence eye condition.
Purpose of the Study:
- To evaluate the diagnostic performance of deep learning (DL) for cataract detection and classification.
- To compare DL accuracy against traditional machine learning and human expert performance.
- To assess DL's potential for automated cataract diagnosis.
Main Methods:
- Systematic literature search across major databases (PubMed, Embase, etc.) up to April 2025.
- Quality assessment of included studies using QUADAS-2.
- Bivariate mixed-effects models for meta-analysis and Deeks' funnel plots for publication bias.
Main Results:
- Sixty-three studies were included; high risk of bias noted in patient selection and index tests.
- Image-based DL achieved high sensitivity (96%) and specificity (98%) for cataract detection (AUC 0.99).
- DL demonstrated strong performance in classification (sensitivity 94%, specificity 97%, AUC 0.99), but lower accuracy on external datasets (e.g., 87% sensitivity).
Conclusions:
- Image-based DL shows high precision for cataract detection and classification, potentially outperforming traditional ML.
- DL performance is comparable to human experts, indicating feasibility for automated diagnosis.
- Limited validation data and generalization challenges warrant caution for broad clinical implementation.
