[论文解读] Weakly Supervised Patch Annotation for Improved Screening of Diabetic Retinopathy

Shramana Dey, Abhirup Banerjee+3 more|arXiv (Cornell University)|Mar 4, 2026

Retinal Imaging and Analysis被引用 0

一句话总结

SAFE：一个两阶段框架，使用弱监督和补丁级嵌入来扩展稀疏的病灶注释于眼底图像，提高 DR 分类性能。

ABSTRACT

Diabetic Retinopathy (DR) requires timely screening to prevent irreversible vision loss. However, its early detection remains a significant challenge since often the subtle pathological manifestations (lesions) get overlooked due to insufficient annotation. Existing literature primarily focuses on image-level supervision, weakly-supervised localization, and clustering-based representation learning, which fail to systematically annotate unlabeled lesion region(s) for refining the dataset. Expert-driven lesion annotation is labor-intensive and often incomplete, limiting the performance of deep learning models. We introduce Similarity-based Annotation via Feature-space Ensemble (SAFE), a two-stage framework that unifies weak supervision, contrastive learning, and patch-wise embedding inference, to systematically expand sparse annotations in the pathology. SAFE preserves fine-grained details of the lesion(s) under partial clinical supervision. In the first stage, a dual-arm Patch Embedding Network learns semantically structured, class-discriminative embeddings from expert annotated patches. Next, an ensemble of independent embedding spaces extrapolates labels to the unannotated regions based on spatial and semantic proximity. An abstention mechanism ensures trade-off between highly reliable annotation and noisy coverage. Experimental results demonstrate reliable separation of healthy and diseased patches, achieving upto 0.9886 accuracy. The annotation generated from SAFE substantially improves downstream tasks such as DR classification, demonstrating a substantial increase in F1-score of the diseased class and a performance gain as high as 0.545 in Area Under the Precision-Recall Curve (AUPRC). Qualitative analysis, with explainability, confirms that SAFE focuses on clinically relevant lesion patterns; and is further validated by ophthalmologists.

研究动机与目标

• 当病灶注释稀缺或不完整时，提升 DR 筛查的动机。
• 提出一种原则性的自动方法，在不进行大量手工标注的情况下扩展稀疏的补丁级注释。
• 通过对 128x128 补丁而非下采样后的图像进行处理，保留病灶级分辨率。
• 通过 refined 注释，使下游任务如 DR 分类获得更高准确度。

提出的方法

• 提出 SAFE，一个将弱监督、对比学习和补丁级嵌入相结合的两阶段框架。
• 阶段 1 训练双臂 Patch Embedding Network (PEN)，从专家标注的补丁中学习类别判别且语义结构化的嵌入。
• 阶段 2 通过在 4 个独立训练的嵌入空间中使用余弦相似度和多数投票（有拒绝）进行特征空间集合，传播未标记补丁的标签。
• 拒绝机制使用置信阈值，在可靠标注与噪声覆盖之间实现平衡。
• 新增度量指标 Drate 和扩展的 MR，用以量化在拒绝下的注释覆盖率和可靠性。
• 实验在四个 DR 数据集上进行补丁级评估，结果显示用 SAFE 推断标签时 DR 分类指标有提升。

实验结果

研究问题

• RQ1 是否可以使用弱监督、补丁级嵌入来可靠地为 DR 图像中的未标记补丁标注？
• RQ2 嵌入空间的集合如何影响标签传播质量以及在部分监督下的鲁棒性？
• RQ3 SAFE 生成的注释对 DR 分类性能的下游影响如何？
• RQ4 拒绝与置信阈值如何影响注释质量与覆盖率？
• RQ5 经 ophthalmologist 验证，推断的注释是否聚焦于临床相关的病灶模式？

主要发现

• SAFE 在所评估数据集上实现 healthy 与 diseased 补丁的可靠分离，准确度最高可达 0.9886。
• SAFE 生成的注释显著提升下游 DR 任务的性能，疾病类别的 F1 分数提高，AUPRC 最高提升至 0.545。
• 两阶段 SAFE 框架结合嵌入空间集合可降低标签传播偏差，并在多个数据集上提升指标。
• 拒绝机制通过将不确定的补丁标注为 Undecided，有效避免了噪声标签，保持 Healthy 与 Unhealthy 类别的精确性。
• 解释性分析（激活图、Grad-CAM）表明 SAFE 聚焦于经 ophthalmologists 验证的临床相关病灶模式。
• 在数据稀缺、类别不平衡或大规模设置下，SAFE 仍保持强劲性能，且注释覆盖率高（Drate > 0.93）。