基于多模型融合方法的肺结节良恶性分类

Multi-model fusion classification method for benign and malignant lung nodules with embedded attention mechanism

针对CT图像中肺结节因边缘模糊、特征不明显造成的分类效果有偏差的问题,本文提出一种嵌入注意力机制的多模型融合方法(简称MSMA-Net)。该方法先将原始CT图像进行肺实质分割和裁剪操作后得到两种不同尺寸的图像,然后分别输入到空间注意力模型和通道注意力模型进行训练,其中,空间注意力模型着重于提取肺结节在CT图像中的空间位置信息,通道注意力模型着重于提取肺结节的细节特征。最后将两个模型提取的特征进行融合,用于得出良恶性分类结果。经过大量实验表明,这种多模型融合方法能很好地提取到肺结节在CT图像中的位置信息和自身的边缘特征,在LIDC数据集的基础上,该方法在准确率,敏感性,特异性分别达到了96.28%,96.72%,96.17%,相较于传统的网络模型取得了更好的分类效果。

Aiming at the problem that the classification effect of lung nodules in CT images is biased due to blurred edges and unobvious features,this paper proposes a multi-model fusion method(MSMA-Net)embedded in the attention mechanism.This method first performs lung parenchymal segmentation and cropping operations on the original CT image to obtain two images of different sizes,and then input them to the spatial attention model and the channel attention model for training.The spatial attention model focuses on extracting lung nodules.The spatial position information of the nodules in CT images,the channel attention model focuses on extracting the detailed features of lung nodules.Finally,the features extracted by the two models are fused to obtain the benign and malignant classification results.A large number of experiments have shown that this multi-model fusion method can well extract the position information of lung nodules in CT images and their own edge features.Based on the LIDC data set,this method is accurate and sensitive.The specificity reached 96.28%,96.72%,and 96.17%,respectively,which achieved better classification results than traditional network models.