Learning to fuse and reconstruct multi-view graphs for diabetic retinopathy grading.
Authors
Affiliations (12)
Affiliations (12)
- Monash University, Wellington Rd, Melbourne, 3800, VIC, Australia. Electronic address: [email protected].
- Shantou University, No. 243, Daxue Road, Shantou, 515063, Guangdong, China.
- University of Wollongong, Wollongong, Northfields Ave, Wollongong, 2500, NSW, Australia.
- Shenzhen University, Nanhai Blvd, Shenzhen, 518060, Guangdong, China.
- Nanjing University of Aeronautics and Astronautics, Yudao Street, Nanjing, 210016, Jiangsu, China.
- University of Queensland, St Lucia, Brisbane, 4072, QLD, Australia.
- University of Sydney, Camperdown, Sydney, 2050, NSW, Australia.
- Griffith University, Kessels Road, Brisbane, 4111, QLD, Australia.
- NVIDIA, San Tomas Expy, Santa Clara, 95051, CA, USA.
- Harbin Institute of Technology (Shenzhen), Pingshan 1 Rd, Shenzhen, 518000, Guangdong, China.
- University of Wollongong, Wollongong, Northfields Ave, Wollongong, 2500, NSW, Australia. Electronic address: [email protected].
- Harbin Institute of Technology (Shenzhen), Pingshan 1 Rd, Shenzhen, 518000, Guangdong, China. Electronic address: [email protected].
Abstract
Diabetic retinopathy (DR) is one of the leading causes of vision loss worldwide, making early and accurate DR grading critical for timely intervention. Recent clinical practices leverage multi-view fundus images for DR detection with a wide coverage of the field of view (FOV), motivating deep learning methods to explore the potential of multi-view learning for DR grading. However, existing methods often overlook the inter-view correlations when fusing multi-view fundus images, failing to fully exploit the inherent consistency across views originating from the same patient. In this work, we present MVGFDR, an end-to-end Multi-View Graph Fusion framework for DR grading. Different from existing methods that directly fuse visual features from multiple views, MVGFDR is equipped with a novel Multi-View Graph Fusion (MVGF) module to explicitly disentangle the shared and view-specific visual features. Specifically, MVGF comprises three key components: ① Multi-view Graph Initialization, which constructs visual graphs via residual-guided connections and employs Discrete Cosine Transform (DCT) coefficients as frequency-domain anchors; ② Multi-view Graph Fusion, which integrates selective nodes across multi-view graphs based on frequency-domain relevance to capture complementary view-specific information; and ③ Masked Cross-view Reconstruction, which leverages masked reconstruction of shared information across views to facilitate view-invariant representation learning. Extensive experimental results on MFIDDR, by far the largest multi-view fundus image dataset, demonstrate that the proposed approach achieves state-of-the-art overall performance for diabetic retinopathy grading. Code is available at https://github.com/HaoranLi525/MVGFDR.