基于三维全卷积网络的肝脏和肝癌分割算法研究

Research on Liver and Liver Tumor Segmentation Algorithm Based on 3D Full Convolutional Network

摘要:

为了解决计算机断层扫描(computed tomography,CT)影像中肝脏和肝癌的准确分割问题，提出了基于三维全卷积网络的肝脏分割算法和肝癌分割算法。肝脏分割算法和肝癌分割算法都采用Vnet网络进行分割。在肝脏分割算法中，采用了形态学方法进行后处理，提高了肝脏分割准确率。在肝癌分割算法中，采用了组合损失函数训练Vnet网络,使得Vnet网络更好地收敛，并加入后处理提高了肝癌分割准确率。为了验证算法的性能，采用MICCAI 2017 Liver Tumor Segmentation Challenge(LiTS)数据集进行了肝脏分割和肝癌分割的5折交叉验证实验。肝脏分割算法在测试集的平均分割准确率为0.9510，高于Unet网络和3D Unet网络；肝癌分割算法的平均分割准确率为0.712。实验结果表明,肝脏分割算法可以准确地对肝脏进行分割，肝癌分割算法也达到了较高的准确率。

Abstract:

In order to solve the problem of accurate segmentation of liver and liver tumor in computed tomography(CT) images, a liver segmentation algorithm based on 3D full convolution network and a liver tumor segmentation algorithm is proposed. Both the liver segmentation algorithm and the liver tumor segmentation algorithm are based on the Vnet network. In the liver segmentation algorithm, the morphological method is used for post-processing, which improves the liver segmentation accuracy. In the liver tumor segmentation algorithm, the combined loss function is used to train the Vnet network, which makes the Vnet network better converge. Post-processing is used to improve the liver tumor segmentation accuracy. In order to verify the performance of the algorithm, a 5-fold cross-validation experiment of liver segmentation and liver tumor segmentation was performed using the MICCAI 2017 Liver Tumor Segmentation Challenge (LiTS) dataset. The average segmentation accuracy of the liver segmentation algorithm in the test set was 0.9510, which was higher than that of the Unet network and the 3D Unet network; the average segmentation accuracy of the liver tumor segmentation algorithm was 0.712. The experimental results show that the liver segmentation algorithm can accurately segment the liver, and the liver tumor segmentation algorithm also achieves a high accuracy.