基于3D＿ResUnet肝脏CT图像分割的临床应用研究

Clinical Application Research of Liver CT Images Segmentation Based on 3D＿ResUnet

目的:为解决传统肝实质分割方法在阈值分割方面存在的分割精度低的问题。方法:采用AI自动识别算法,通过Unet与Resnet相结合的3D_ResUnet网络对肝脏CT图像进行分割,并对分割结果通过最大联通分量的方法去除杂质,得到较为精确的肝脏区域,实现肝实质自动分割。结果:基于3D_ResUnet的肝脏CT图像分割,其分割的平均Dice为96.12%,高于3D_Unet的分割精度。结论:基于3D_ResUnet的肝脏CT图像分割提高了肝实质分割的精度,实现了无需人工交互的全自动分割,通过应用在肝癌手术计划系统中,为临床医生的肝癌手术规划提供了可视化依据。

Objective: To solve the problem of low segmentation accuracy in threshold segmentation of traditional liver parenchyma segmentation methods. Methods: Using AI Automatic recognition Algorithm, through 3 D_ResUnet network combined by Unet and Resnet, to make the segmentation on the liver CT images, then removed the impurities from segmentation result by maximum connected component method, to obtain a more accurate liver region and finally realize the liver parenchyma segmentation. Results: The average dice of liver CT images segmentation based on 3 D_ResUnet is 96.12%, which is higher than the segmentation accuracy of 3 D_Unet. Conclusion: Liver CT image segmentation based on 3 D_ResUnet improves the accuracy of liver parenchymal segmentation, the fully automatic segmentation that without human interaction is realized, though the application in liver cancer plan system, to provide a visual basis of liver cancer surgeon plan for clinical doctors.