基于改进Unet网络的CT图像肺实质自动分割算法

AUTOMATIC SEGMENTATION ALGORITHM OF LUNG PARENCHYMAIN CT IMAGES BASED ON IMPROVED UNET NETWORK

肺实质分割是计算机辅助诊断肺癌中的重要步骤。针对Unet分割精度不足和收敛速度慢的问题,提出一种基于改进Unet的肺实质分割算法。采用K-means聚类和凸包扫描算法进行预分割,完成肺实质的定位和修正。以Unet结构为基础,引入Sobel卷积层强化边缘区域的高通滤波,并在特征融合中加入随机失活模块,进一步提升分割精度。将传统图像处理方法与深度学习相结合,获得了优化改进的分割模型。实验表明,该算法可以准确高效地分割肺实质,平均Dice相似系数达到0.9834,收敛速度和分割性能均优于其他几种较新的分割算法。

Segmentation of lung parenchyma is an important step in computer-aided diagnosis of lung cancer.Aimed at the problems of insufficient segmentation accuracy and slow convergence speed of Unet,a lung parenchymal segmentation algorithm based on improved Unet is proposed.K-means clustering and convex hull scanning algorithm were used for pre-segmentation to complete the positioning and correction of lung parenchyma.Based on the Unet structure,the Sobel convolutional layer was introduced to strengthen the high-pass filtering of the edge area,and the random inactivation module was added to the feature fusion to further improve the segmentation accuracy.Combining traditional image processing methods with deep learning,an optimized and improved segmentation model was obtained.Experiments show that the algorithm can segment lung parenchyma accurately and efficiently,with an average Dice similarity coefficient of 0.9834,and the convergence speed and segmentation performance are better than other new segmentation algorithms.