Explainable Fundus Image Curation and Lesion Detection in Diabetic Retinopathy

2025-12-06

AI Summary

Early diagnosis of diabetic retinopathy (DR) relies on high-quality fundus images and precise lesion annotations; however, clinical datasets often suffer from substantial annotation noise and inconsistent labeling standards, severely limiting AI model performance. To address this, we propose an interpretable framework for fundus image quality control and lesion detection tailored to DR. Our method introduces a novel interpretable feature classifier that integrates image processing with contrastive learning for fine-grained image quality assessment, and designs a multi-annotator agreement evaluation mechanism grounded in a consistency formulation to quantify and enhance annotation reliability. The framework unifies interpretable classification, deep learning–assisted annotation, and multi-source consistency modeling. Evaluated on public DR datasets, it achieves a 23.5% improvement in inter-annotator agreement and a 31% reduction in mislabeling rate, significantly enhancing the robustness and clinical trustworthiness of downstream lesion detection models.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Diabetic Retinopathy (DR) affects individuals with long-term diabetes. Without early diagnosis, DR can lead to vision loss. Fundus photography captures the structure of the retina along with abnormalities indicative of the stage of the disease. Artificial Intelligence (AI) can support clinicians in identifying these lesions, reducing manual workload, but models require high-quality annotated datasets. Due to the complexity of retinal structures, errors in image acquisition and lesion interpretation of manual annotators can occur. We proposed a quality-control framework, ensuring only high-standard data is used for evaluation and AI training. First, an explainable feature-based classifier is used to filter inadequate images. The features are extracted both using image processing and contrastive learning. Then, the images are enhanced and put subject to annotation, using deep-learning-based assistance. Lastly, the agreement between annotators calculated using derived formulas determines the usability of the annotations.

Problem

• Develops an explainable quality-control framework for fundus images
• Filters inadequate images using feature-based classifiers and contrastive learning
• Enhances images and assesses annotation agreement for AI training

Innovation

• Explainable feature-based classifier filters inadequate images
• Image enhancement and deep-learning-assisted annotation process
• Annotator agreement formulas determine annotation usability