FundaQ-8: A Clinically-Inspired Scoring Framework for Automated Fundus Image Quality Assessment

Automated fundus image quality assessment (FIQA) remains challenging due to variations in image acquisition and subjective expert evaluations. This study introduces FundaQ-8, a novel, expert-validated framework designed to systematically assess fundus image quality based on eight critical parameters, including field coverage, anatomical visibility, and others. Using FundaQ-8 as a structured scoring reference, a ResNet18-based regression model is built to predict continuous quality scores (0-1), trained on 1,800 fundus images from real-world clinical and Kaggle datasets. The model leverages transfer learning with an MSE-optimized regression adaptation, and standardized preprocessing. Validation against the EyeQ dataset and statistical analyses confirm the framework’s reliability in providing interpretable, clinically meaningful assessments. Furthermore, our findings indicate that incorporating FundaQ-8 as the scoring framework for FIQA into deep learning-based diabetic retinopathy (DR) grading models enhances diagnostic robustness.