Lung Tumor Segmentation Algorithm Based on Deep Convolution Neural Network
Download as PDF
CHEN Shu, JI Jianbing, YANG Yuanyuan
In order to segment lung tumor in CT image automatically, a segmentation method based on improved 3DUnet is proposed. The convolution kernel size is increased, PReLU is used as the activation function, the training set is expanded and optimized by nonlinear transformation, the network is trained by DICE loss function and SGDM gradient descent algorithm, post-processing is added to optimize the segmentation results. The experiment is carried out by using medical segmentation decathlon open data set and 5 indicators are used for evaluation. Results show that the algorithm achieves high accuracy and is better than 3DUnet network.
CNN, Lung tumor, Segmentation