Semi-Supervised Multi-Task Learning for Interpretable Quality As- sessment of Fundus Images

2025-11-17

AI Summary

Existing retinal image quality assessment (RIQA) methods typically output only a global quality label, lacking fine-grained localization of specific acquisition artifacts—such as uneven illumination, blur, or low contrast—thus hindering clinical guidance for image reacquisition; this limitation stems primarily from the prohibitively high cost of fine-grained annotation. Method: We propose a semi-supervised multi-task learning framework built upon ResNet-18, integrating a teacher–student architecture to generate high-fidelity pseudo-labels. The model jointly predicts both global quality grades and multiple imaging defect types without requiring additional human annotations. Contribution/Results: Our approach significantly enhances model interpretability and clinical utility. On EyeQ and DeepDRiD benchmarks, it achieves F1-scores of 0.875 and 0.778, respectively. Fine-grained defect identification performance is statistically indistinguishable from the teacher model (p > 0.05), and on a newly annotated subset, it approaches expert-level accuracy.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Knowledge Representation and Reasoning: Qualitative Reasoning

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

Abstract

Retinal image quality assessment (RIQA) supports computer-aided diagnosis of eye diseases. However, most tools classify only overall image quality, without indicating acquisition defects to guide recapture. This gap is mainly due to the high cost of detailed annotations. In this paper, we aim to mitigate this limitation by introducing a hybrid semi-supervised learning approach that combines manual labels for overall quality with pseudo-labels of quality details within a multi-task framework. Our objective is to obtain more interpretable RIQA models without requiring extensive manual labeling. Pseudo-labels are generated by a Teacher model trained on a small dataset and then used to fine-tune a pre-trained model in a multi-task setting. Using a ResNet-18 backbone, we show that these weak annotations improve quality assessment over single-task baselines (F1: 0.875 vs. 0.863 on EyeQ, and 0.778 vs. 0.763 on DeepDRiD), matching or surpassing existing methods. The multi-task model achieved performance statistically comparable to the Teacher for most detail prediction tasks (p > 0.05). In a newly annotated EyeQ subset released with this paper, our model performed similarly to experts, suggesting that pseudo-label noise aligns with expert variability. Our main finding is that the proposed semi-supervised approach not only improves overall quality assessment but also provides interpretable feedback on capture conditions (illumination, clarity, contrast). This enhances interpretability at no extra manual labeling cost and offers clinically actionable outputs to guide image recapture.

Problem

• Developing semi-supervised multi-task learning for retinal image quality assessment
• Providing interpretable feedback on acquisition defects without extensive manual labeling
• Enhancing clinically actionable outputs to guide fundus image recapture

Innovation

• Semi-supervised learning with pseudo-labels for quality assessment
• Multi-task framework combining overall and detailed quality predictions
• Teacher model generates weak annotations to reduce manual labeling