A New Strategy for Retinal Vessel Segmentation Based on Semi-Dilation Supervised Learning and Boundary Reconstruction

Automatic segmentation of retinal blood vessels is an important means to assist doctors in quickly diagnosing and preventing eye diseases. However, due to the scale differences between blood vessels and the complexity of blood vessel morphology, the development of automatic retinal vessel segmentation is hindered. In response to these segmentation difficulties, this paper proposes a new method based on semi-dilation supervised learning and vessel boundary reconstruction. By using semi-dilation supervised learning during the pre-training stage, the scale differences between blood vessels are effectively alleviated, and the convolution kernel's perception range of fine blood vessels is expanded. In formal training, the boundary reconstruction network removes redundant blood vessels introduced in pre-training and enhances the ability to determine blood vessel boundaries. Our proposed method was ultimately tested on the DRIVE dataset. The results show that the proposed method achieves good results in segmenting fine blood vessels, with a sensitivity of 84.65%, and the overall segmentation effect is improved, with an AUC of 98.84%. The proposed method effectively solves the problem of ignoring subtle blood vessels caused by scale differences.