基于Attention U-Net的肺结节自动检测优化

Optimizing Pulmonary Nodule Detection with Attention U-Net

摘要:

针对胸部CT影像中肺结节检测的精度需求，本研究提出了一种基于注意力机制的U-Net网络（Attention U-Net）。该网络结合了U-Net的编码-解码架构及注意力机制，以增强重要特征的提取和不相关信息的抑制，从而提高结节检测的精度。在网络中，注意力模块有助于在多尺度特征图中自适应地加权，使模型能够更好地集中于肺结节的区域。我们在公共数据集LUNA 16上进行了一系列实验，结果表明，使用Attention U-Net网络的检测性能明显优于传统方法，且提供了更高的准确性和召回率。此外，在临床数据集TCIA上，实验结果进一步验证了该网络在实际应用中的有效性和可靠性。这表明基于Attention U-Net的肺结节检测方法在胸部CT扫描的临床实践中具有广泛的应用前景。

Abstract:

To address the precision requirements for pulmonary nodule detection in chest CT images, this study proposes an attention-enhanced U-Net network (Attention U-Net). The network integrates an encoder-decoder architecture with attention mechanisms, enhancing the extraction of critical features while suppressing irrelevant information to improve detection accuracy. The attention module adaptively weights multi-scale feature maps, enabling the model to focus on nodule regions and mitigate interference from noise. Experimental evaluations on the public LUNA16 dataset demonstrate that the proposed Attention U-Net significantly outperforms traditional methods, achieving higher accuracy and recall rates. Further validation on the clinical TCIA dataset confirms the network’s effectiveness and reliability in real-world applications. These results highlight the broad clinical potential of the Attention U-Net-based approach for pulmonary nodule detection in chest CT screening.