Fig 1.
The encoder and decoder of LM-UNet consist of PV-Mamba+EMA, with skip connections formed by EFF and MMSC. EFE is employed to fuse edge features from different levels of encoder layers and
.
Fig 2.
Parallel PV-Mamba Architecture.
Fig 3.
Efficient Multi-scale Attention.
Fig 4.
The Edge Feature Extraction Architecture.
Fig 5.
The Edge Feature Fusion Architecture.
Fig 6.
Multi-stage Multi-level Skip Connections.
Table 1.
Detailed Comparison of LM-UNet and UVM-UNet.
Table 2.
Detailed Information on Data Collection for Each Center.
Table 3.
LM-UNet encoder input and output dimensions.
Table 4.
Performance comparison of different methods.
Fig 7.
Segmentation Results of Different Segmentation Methods on PROMISE12 Dataset.
Fig 8.
Bar Chart of Evaluation Metrics for Different Segmentation Methods.
Table 5.
Ablation experiments of different components.
Table 6.
Complexity analysis of different methods.
Fig 9.
Complexity analysis of different methods.