一种结合注意力残差的肝脏及肝肿瘤分割算法

SEGMENTATION OF LIVER AND LIVER TUMOR BASED ON ATTENTION RESIDUAL

长时间的肝脏医学图像人工诊断容易使医生产生疲劳,导致误诊和漏诊情况发生。针对以上现象提出一种改进的Unet网络用于肝脏和肝肿瘤自动分割。改进Unet模型,引入注意力残差结构和特征复用结构,提高输入图像中特征信息的利用效率;对损失函数进行改进,在Dice系数中加入欠分割和过分割惩罚因子,提高模型的预测能力。在公开数据集上的实验结果表明:该算法对肝脏和肝肿瘤的分割相似系数分别达到了0.962和0.713,优于现有的分割模型且具有较强的鲁棒性。

The long-term manual diagnosis of liver medical images is likely to cause fatigue to the doctor,leading to misdiagnosis and missed diagnosis.Aimed at the above phenomenon,an improved Unet network is proposed for automatic segmentation of liver and liver tumors.We improved the Unet model amd introduced the attention residual structure and feature multiplexing structure to improved the utilization efficiency of the feature information in the input image.We improved the loss function and added the under-segmentation and over-segmentation penalty factors to the Dice coefficient to improve the model s predictive ability.The experimental results on the public dataset show that the segmentation similarity coefficients of the algorithm in the liver and liver tumors reaches 0.962 and 0.713,respectively,which is better than the existing segmentation models and has strong robustness.