EyeBench: A Call for More Rigorous Evaluation of Retinal Image Enhancement

• DOI: 10.48550/arxiv.2502.14260
• Publication Date: 2025-02-19
• Authors: Wenhui Zhu; Xuanzhao Dong; Xin Li; Yujian Xiong; Xiwen Chen; Peijie Qiu; Vamsi Krishna Vasa; Zhangsihao Yang; Su Yi; Oana Dumitrașcu; Yalin Wang
• Keywords: FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV); Computer Science - Computer Vision and Pattern Recognition; FOS: Electrical engineering, electronic engineering, information engineering; Electrical Engineering and Systems Science - Image and Video Processing

Over the past decade, generative models have achieved significant success in enhancement fundus images.However, evaluation of these still presents a considerable challenge. A comprehensive benchmark for image is indispensable three main reasons: 1) The existing denoising metrics (e.g., PSNR, SSIM) are hardly to extend downstream real-world clinical research Vessel morphology consistency). 2) There lack both paired and unpaired methods, along with need expert protocols accurately assess value. 3) An ideal system should provide insights inform future developments enhancement. To this end, we propose novel benchmark, EyeBench, that align needs, offering foundation work improve relevance applicability EyeBench has appealing properties: multi-dimensional alignment evaluation: In addition evaluating task, several clinically tasks images, including vessel segmentation, DR grading, generalization, lesion segmentation. Medical expert-guided design: We introduce dataset promote fair comparisons between methods includes manual protocol by medical experts. Valuable insights: Our study provides rigorous across different tasks, assisting experts making informed choices. Additionally, offer further analysis challenges faced methods. code available at \url{https://github.com/Retinal-Research/EyeBench}