基于改进U-Net的肝脏分割方法

针对现有肝脏图像分割方法存在分割精度较低的问题，提出一种改进U-Net的肝脏分割方法。该方法对U-Net结构做出以下改进，即引入改进的残差模块、重新设计跳跃连接，然后采用混合损失函数，从而提高特征信息的利用率，减少编码器和解码器之间的语义差异，缓解类不平衡的问题并且加快网络收敛。在CodaLab组织提供的公共数据集LITS（Liver Tumor Segmentation）上的实验结果表明，利用该方法达到的Dice相似系数值、敏感度、交并比分别为93.69%、94.87%和87.49%。相比于U-Net和Attention U-Net等分割方法，该方法分割出的肝脏区域结果更加准确，取得了更好的分割性能。

Liver segmentation method based on improved U-Net

Abstract: In order to solve the problem of low precision in existing methods for liver segmentation, a liver segmentation method based on improved U-Net is proposed. U-Net structure is improved by introducing improved residual block and redesigning skip connection, and then mixed loss function is adopted to enhance the utilization of feature information and reduce the semantic differences between encoder and decoder, thereby alleviating class imbalance problem and speeding up network convergence. The experimental results on Liver Tumor Segmentation (LITS), a common data set provided by CodaLab, showed that the Dice similarity coefficient, sensitivity and intersection over union achieved by the proposed method were 93.69%, 94.87% and 87.49%, respectively. Compared with other segmentation methods, such as U-Net and Attention U-Net, the proposed method can obtain a more accurate result in liver segmentation and has better segmentation performance.