To develop an effective lung nodule segmentation technique that integrates CNN and Transformer architectures for improved accuracy in CT images.
Key Findings:
DPCrossU-Net achieved a Dice score of 85.89% on the LIDC-IDRI dataset.
It outperformed the baseline U-Net, especially in scenarios involving small nodules and complex backgrounds.
Interpretation:
The results suggest that combining CNN and Transformer feature extraction with adaptive cross-branch fusion enhances lung nodule segmentation accuracy.
Limitations:
The study does not address the computational efficiency of DPCrossU-Net compared to existing models.
Further validation on diverse datasets is necessary to confirm generalizability.
Conclusion:
DPCrossU-Net offers a robust solution for lung nodule segmentation, potentially aiding early lung cancer detection and future diagnostic systems.
