A refined equilibrium generative adversarial network for retinal vessel segmentation

Objective: Retinal vessel morphological parameters are vital indicator for early diagnosis of ophthalmological diseases and cardiovascular events. However, segmentation performance is highly influenced by elusive vessels, especially in low-contrast background and lesion regions. In this work, we present an end-to-end synthetic neural network to strengthen elusive vessels segmentation capability, containing a symmetric equilibrium generative adversarial network (SEGAN), multi-scale features refine blocks (MSFRB), and attention mechanism (AM). Method: The proposed network is superior in detail information extraction by maximizing multi-scale features representation. First, SEGAN constructs a symmetric adversarial architecture in which generator is forced to produce more realistic images with local details. Second, MSFRB are devised to optimize the feature merging process, thereby maximally maintaining high resolution information. Finally, the AM is employed to encourage the network to concentrate on discriminative features. Results: On public dataset DRIVE, STARE, CHASEDB1, and HRF, we evaluate our network quantitatively and compare it with state-of-the-art works. The ablation experiment shows that SEGAN, MSFRB, and AM both contribute to the desirable performance. Conclusion: The proposed network outperforms the existing methods and effectively functions in elusive vessels segmentation, achieving highest scores in Sensitivity, G-Mean, Precision, and F1-Score while maintaining the top level in other metrics. Significance: The satisfactory performance and computational efficiency offer great potential in clinical retinal vessel segmentation application. Meanwhile, the network could be utilized to extract detail information in other biomedical image computing. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

This study presents an end-to-end synthetic neural network to enhance segmentation of elusive vessels, achieving satisfactory results by maximizing multi-scale feature representation. SEGAN, MSFRB, and AM all contribute to the network's performance improvements according to various evaluations.