Adaptive Class Learning to Screen Diabetic Disorders in Fundus Images of Eye

2025-01-21

International Conference on Pattern Recognition

AI Summary

To address the challenges of scarce labeled data, severe class imbalance, and label noise in fine-grained ocular disease recognition (e.g., diabetic retinopathy) from fundus images, this paper proposes an adaptive class-aware robust classification framework. Methodologically, it integrates a dynamic class-weighted reweighting mechanism with an uncertainty-aware pseudo-label optimization strategy, built upon a ResNet backbone, an adaptive loss function, curriculum learning scheduling, consistency regularization, and Monte Carlo Dropout for uncertainty estimation—all without requiring additional annotations. On the APTOS and EyePACS benchmarks, the method achieves F1-scores of 89.7% and 87.3%, respectively, surpassing state-of-the-art methods by 3.2% and demonstrating significantly improved robustness against severe label noise.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Problem

• Diabetic Retinopathy Detection
• Limited Data Training
• Ophthalmic Disease Diagnosis

Innovation

• CELD Method
• Limited Data Learning
• Disease Classification in Ophthalmology