A 3D SCDA U-Net Algorithm for Lung Parenchyma Segmentation Based on Dual Attention Mechanism

Lung parenchyma segmentation is vital for lung disease diagnosis. The traditional U-Net suffers from insufficient feature fusion in skip connections, causing detail loss. This paper presents a 3D SCDA U-Net algorithm with dual attention mechanism. By integrating channel and position attention modules, it enhances feature fusion and global context modeling. Experiments on the LUNA16 dataset show it achieves Dice and Jaccard coefficients of 98.22% and 97.09%, outperforming existing mainstream algorithms. Ablation studies confirm the dual attention modules’ effectiveness in reducing over-segmentation and under-segmentation issues.