Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
97 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Attention Guided Network for Retinal Image Segmentation (1907.12930v3)

Published 25 Jul 2019 in eess.IV and cs.CV

Abstract: Learning structural information is critical for producing an ideal result in retinal image segmentation. Recently, convolutional neural networks have shown a powerful ability to extract effective representations. However, convolutional and pooling operations filter out some useful structural information. In this paper, we propose an Attention Guided Network (AG-Net) to preserve the structural information and guide the expanding operation. In our AG-Net, the guided filter is exploited as a structure sensitive expanding path to transfer structural information from previous feature maps, and an attention block is introduced to exclude the noise and reduce the negative influence of background further. The extensive experiments on two retinal image segmentation tasks (i.e., blood vessel segmentation, optic disc and cup segmentation) demonstrate the effectiveness of our proposed method.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Shihao Zhang (44 papers)
  2. Huazhu Fu (185 papers)
  3. Yuguang Yan (14 papers)
  4. Yubing Zhang (2 papers)
  5. Qingyao Wu (39 papers)
  6. Ming Yang (289 papers)
  7. Mingkui Tan (124 papers)
  8. Yanwu Xu (78 papers)
Citations (171)

Summary

Attention Guided Network for Retinal Image Segmentation

The paper introduces an innovative methodology for retinal image segmentation, specifically aimed at improving the automatic diagnosis of eye diseases by precisely segmenting structures like blood vessels, optic discs, and cups. The authors present the Attention Guided Network (AG-Net), which addresses the limitations of traditional Convolutional Neural Networks (CNNs) in preserving structural information during segmentation tasks.

Key Contributions

AG-Net utilizes a guided filter combined with an attention mechanism to enhance the segmentation process. The novel approach incorporates this attention-guided filter into the expanding path of the network, allowing for greater retention of structural details critical to the task. Unlike previous models such as FCN and U-Net, which employ simple skip connections for feature fusion, AG-Net integrates an edge-preserving guided filter that transfers low-level structural features to higher levels efficiently.

The attention block within the filter mechanism plays a significant role by pinpointing meaningful foreground elements while mitigating background noise, ensuring noise-free and clear segmentation.

Experimental Evaluation

The paper provides substantial evidence of AG-Net's efficacy through rigorous testing on two prevalent retinal image datasets—DRIVE and ORIGA. In the DRIVE dataset, which focuses on vessel segmentation, AG-Net shows improved accuracy and Intersection-over-Union (IOU), outperforming state-of-the-art methods by leveraging its attention-guided mechanism, which enriches discriminative power and improves structural feature preservation. Key metrics such as AUC and sensitivity further reflect the superiority of AG-Net over traditional CNN methods.

For optic disc and cup segmentation on the ORIGA dataset, AG-Net demonstrates lower overlapping error compared to existing methodologies. Incorporating polar transformation further enhances the segmentation accuracy, which illustrates AG-Net's potential when extended with additional transformations.

Implications and Future Scope

This paper substantially contributes to the field of medical imaging, specifically in the domain of retinal image analysis. The approach proposed offers a feasible solution to the inherent problem of losing critical structural information with traditional pooling operations in CNNs. By providing a mechanism for retaining and enhancing these details, AG-Net could lead to more accurate diagnoses and potentially assist in early detection of eye diseases.

In terms of future developments, there are promising avenues for AG-Net in broader medical imaging applications where preserving structural details is crucial. Furthermore, extending the guided attention mechanism to combine other multi-modal data sources could offer comprehensive insights into complex clinical scenarios.

Overall, this research not only advances segmentation techniques within retina imaging but also opens the door for exploration into new domains requiring precise boundary delineation. As AI technologies evolve, such advancements are critical for translating complex data into actionable clinical interventions.