结合线性迭代聚类和分割对抗的肺结节分割

Linear Iterative Clustering Combining Segmentation Adversarial Network for Pulmonary Nodule Segmentation

由于肺结节的形状和大小不一,在进行肺结节语义分割的过程中普遍存在漏分割和交叉比偏低的问题,为了改善这种现象,提出了一种将改进的线性迭代聚类与引入注意力机制的条件分割对抗网络相结合的肺结节分割方法,首先利用SLIC超像素分割法对肺部图像进行预分割,再将预分割好的图像输入以改进的U-Net为分割器的分割对抗网络模型,通过分割器与判别器相互对抗学习优化模型参数以获得更好的肺结节分割效果。实验结果表明,相比与U-Net,本文所用方法Dice系数提升了1.9%,交叉比提升了2.8%。

Due to the different shapes and sizes of pulmonary nodules,there are many problems such as missing segmentation and low recall rate in the process of semantic segmentation of pulmonary nodules.Targeting at improving this phenomenon,an improved deep learning method is proposed to segment pulmonary nodules,which combines the improved linear iterative clustering with the conditional segmentation adversarial network with attention mechanism.Firstly,the SLIC superpixel segmentation method is used to pre-segment the lung image,and then the pre-segmented image is input into the segmentation countermeasure network model with the improved U-Net as the segmentor.Through the mutual countermeasure between the segmentor and the discriminator,the model parameters are optimized to obtain better lung nodule segmentation effect.The simulation results show that compared with U-Net,the Dice coefficient of the method used in this paper is increased by 1.9%,and the recall rate is increased by 2.8%.