A refined equilibrium generative adversarial network for retinal vessel segmentation

Objective: Recognizing retinal vessel abnormity is vital to early diagnosis of ophthalmological diseases and cardiovascular events. However, segmentation results are highly influenced by elusive vessels, especially in low-contrast background lesion region. In this work, we present an end-to-end synthetic neural network, containing a symmetric equilibrium generative adversarial network (SEGAN), multi-scale features refine blocks (MSFRB), attention mechanism (AM) enhance the performance on segmentation. Method: The proposed granted powerful representation capability extract detail information. First, SEGAN constructs architecture, which forces generator produce more realistic images with local details. Second, MSFRB devised prevent high-resolution from being obscured, thereby merging better. Finally, AM employed encourage concentrate discriminative features. Results: On public dataset DRIVE, STARE, CHASEDB1, HRF, evaluate our quantitatively compare it state-of-the-art works. ablation experiment shows that SEGAN, MSFRB, both contribute desirable performance. Conclusion: outperforms mature methods effectively functions vessels segmentation, achieving highest scores Sensitivity, G-Mean, Precision, F1-Score while maintaining top level other metrics. Significance: appreciable computational efficiency offer great potential clinical application. Meanwhile, could be utilized information biomedical issues