PaeNet: parallel attention enhancement network for ophthalmic image segmentation

Image segmentation plays a critical role in medical image analysis. In recent years, U-Net-based semantic segmentation networks have gained wide popularity in various medical image segmentation tasks, including optic disc and optic cup segmentation, corneal nerve segmentation, and blood vessel detection. However, the existing segmentation methods often miss many details and require more robust feature extraction capabilities. Additionally, the continuous convolution and pooling operations result in the loss of spatial information, hindering more accurate segmentation of the target region. To address these issues, a novel Parallel Attention Enhancement Network (PaeNet) is proposed. Specifically, PaeNet introduces a dual-scale feature enhancement attention module that enhances the contextual information by fusing the feature information between adjacent feature layers and reconstructing the skip connections. Furthermore, PaeNet integrates a multi-scale parallel weight-shared fusion module that employs dense dilated convolution to capture deeper semantic information. To extract clear target boundary information, we introduce a new multi-label loss function to improve the ratio of foreground and background pixels. Finally, experiments are conducted on several public optic image datasets to evaluate the performance of PaeNet. The comprehensive results demonstrate that PaeNet outperforms state-of-the-art methods in the optic disc and cup segmentation, corneal nerve segmentation, and blood vessel detection.