To systematically assess the performance of different deep learning models in cataract detection and classification, with a focus on sensitivity, specificity, and area under the ROC curve (AUC), emphasizing the systematic review process.
Key Findings:
Cataracts are a leading cause of visual impairment globally, particularly affecting populations in low- and middle-income countries, with specific statistics on prevalence.
Deep learning models, especially convolutional neural networks (CNNs), show promise in automating cataract detection and classification, with examples of successful implementations.
Existing studies on DL for cataracts exhibit heterogeneous findings, necessitating a systematic review to clarify these discrepancies.
Interpretation:
Deep learning has the potential to enhance cataract diagnosis through improved accuracy and efficiency, particularly in resource-limited settings, by utilizing advanced image processing techniques.
Limitations:
Current DL models often lack compatibility with low-cost equipment used in low-resource settings, highlighting the need for adaptable solutions.
There is a significant decline in model performance in real-world scenarios due to lack of generalization, suggesting the need for diverse training datasets.
Clinical validation of DL models is insufficient, with only 6% of studies conducting external validation, indicating a critical gap in the research.
Conclusion:
The study aims to provide a comprehensive evaluation of DL models for cataract diagnosis, addressing methodological quality and supporting clinical translation, with implications for improving patient outcomes.
