Improving U-net Model for Pulmonary Nodule Segmentation Through Attention Mechanism and Post-processing Module

Pulmonary nodule segmentation plays a crucial role in the early detection and diagnosis of lung cancer, significantly impacting patient outcomes. The U-net model has emerged as a useful architecture in medical image processing like pulmonary nodule segmentation, gaining widespread popularity. However, U-net suffers from poor fine segmentation capabilities and the presence of noise in the segmentation results. In this study, we propose an Enhanced U-net model to improve segmentation results. The Enhanced U-net uses ECA and CRF modules. ECA module can make the model focus more on important features and improve the fine segmentation ability of the model. Meanwhile, CRF module allows the model to further refine the results, reduce the noise and boundary discontinuities. Utilizing the LIDC dataset, we evaluate the models performance through indicators such as recall, IoU, Dice score. Our findings show that Enhanced U-net can achieve the best performance among all U-net based models. And Enhanced U-net can significantly improve segmentation outcomes for small or blurred nodules.