Interlace mixed net of lung nodule segmentation based on Transformer and CNN
Aiming at the problems of multi-size and high heterogeneity of lung nodules,an interlace mixed network based Transformer and convolutional neural network interlace mixed(IMTC)is proposed.The network is a symmetrical hierarchical connection network with strong multi-scale feature extraction capabilities.It integrates two new schemes to solves the promblems of multi-size and shape heterogeneity.① Inception attention module(IAM)is proposed to capture richer shallow features by paralleling multiple convolution kernels of different sizes to increase the combination of receptive fields.② In order to extract deeper semantic features with more expressive ability,the basic backbone network composed of Transformer and CNN is used to extract nodule features alternately,so that the global features and local features are fully integrated,and then the generalization ability and robustness of nodule feature representation are improved.The experimental results show that the model in this paper can accurately segment nodules with small scale and complex margin,and has good segmentation performance on the LUNA16 public dataset,and the Dice and IOU reach 86.15％and 76.10％,respectively.
万方数据
维普
