To provide a comparative evaluation of eight widely adopted transfer learning models for diagnosing ocular disorders, emphasizing both disease-specific performance and interpretability.
Approach:
Key Findings:
Different transfer learning models show varying performance across eye diseases, impacting clinical choices.
Explainable AI methods yield insights that enhance understanding beyond accuracy metrics.
Cross-model consistency indicates a need for segmentation-based models for effective glaucoma diagnosis.
Interpretation:
The study highlights the critical role of explainability in AI models for ocular disorder diagnosis, suggesting that enhanced interpretability can significantly improve clinical decision-making.
Limitations:
The study does not propose a new methodology but evaluates existing models, limiting innovation.
Generalizability is constrained by the specific datasets utilized in the analysis.
Conclusion:
The findings emphasize the necessity of explainable AI in ocular disorder diagnosis, which can lead to improved understanding and decision-making in clinical environments.
