In order to overcome the loss of detail information caused by down sampling of traditional liver tumor segmentation networks and extract rich multi-scale information at the same time, this paper proposes an algorithm of liver tumor segmentation based on dilated convolution of stacked tree aggregation structure. First, a residual dense module is proposed in the encoder network. Then, a dilated convolution module of stacked tree aggregation structure is added to the encoder-decoder network, which can effectively eliminate the checkerboard artifacts caused by ordinary dilated convolution and improve the segmentation accuracy. Finally, a weighted loss function is used to solve the problem of the imbalance between the foreground and the background in the image. The experimental results show that the Dice similarity coefficient, pixel accuracy rate and intersection ratio of the algorithm on the computer tomography image data set are 0.8026, 0.7974 and 0.7317, respectively.