This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
UOPSL: Unpaired OCT Predilection Sites Learning for Fundus Image Diagnosis Augmentation
Created by
Haebom
Author
Zhihao Zhao, Yinzheng Zhao, Junjie Yang, Xiangtong Yao, Quanmin Liang, Daniel Zapp, Kai Huang, Nassir Navab, M. Ali Nasseri
Outline
This paper presents a novel approach to address the challenges of acquiring multimodal images despite significant advancements in the use of multimodal medical imaging in ophthalmic disease diagnosis. Specifically, to address the imbalance between relatively inexpensive and readily available fundus photographs and expensive OCT images, we propose an unpaired multimodal framework, \UOPSL. \UOPSL utilizes spatial prior information (predilection sites) obtained from OCT images to enhance fundus image-based disease recognition. Through contrastive learning on large-scale unpaired OCT and fundus images, \UOPSL learns lesion location patterns in the OCT latent space and utilizes this information to perform disease classification solely on fundus images. We report that our approach outperforms existing methods on nine diverse datasets encompassing 28 key categories.
Takeaways, Limitations
•
Takeaways:
◦
We present a novel unpaired multimodal framework that effectively addresses the modal imbalance problem between fundus photographs and OCT images.
◦
Improving fundus photograph-based disease diagnosis performance by utilizing spatial prior information in OCT images.
◦
Demonstrated superior performance compared to existing methods on various datasets.
•
Limitations:
◦
Further verification of the generalization performance of the proposed \UOPSL framework is needed.
◦
Applicability and limitations for various ophthalmic diseases need to be confirmed.
◦
Further analysis is needed on the reliability and accuracy of spatial prior information extracted from OCT images.
◦
Limited applicability to small datasets due to dependence on large datasets.
