摘要
目的基于胎盘MRI图像,构建Trans-UNet深度学习胎盘自动分割模型,自动测量胎盘体积,并与公式法比较,为提高胎盘植入性疾病(placenta accreta spectrum disorders,PAS)的诊断效能提供可定量参考指标。方法回顾性收集230例于本院行胎盘MRI检查的孕妇的影像学资料,建立Trans-UNet深度学习模型自动分割胎盘并测量胎盘体积。以3Dslicer软件测量的胎盘体积为金标准,比较公式法、3Dslicer、胎盘自动分割计算机测量的胎盘体积的准确性。结果Trans-UNet深度学习模型的IoU值为0.755,Dice系数为0.858。公式法与3Dslicer以及胎盘自动分割计算机测量的胎盘体积有统计学差异,一致性均较差(ICC值为0.232、0.186);胎盘自动分割计算机测量与3Dslicer测量的胎盘体积无统计学差异,一致性极好(ICC值为0.915)。结论本研究建立的Trans-UNet深度学习模型可用于胎盘自动分割,且测量的胎盘体积快速高效、准确性高,为实现胎盘体积自动测量奠定基础。
Objective To develop a Trans-UNet deep learning model for automated placental segmentation on MRI images,enabling automatic placental volume measurement.This aims to provide a quantitative reference for improving the diagnostic efficacy of placenta accreta spectrum disorders(PAS).Methods MRI imaging from 230 pregnant women who underwent placental MRI before delivery at our institution were retrospectively analyzed.A Trans-UNet model was developed to automate placental segmentation and volume calculation.The accuracy of placental volume measurements derived from the formula-based method,3D Slicer software(gold standard),and the automated segmentation model were compared.Results The Trans-UNet deep learning model achieves an IoU of 0.755 and a Dice coefficient of 0.858.Significant differences and poor agreement were observed between formula-based measurements and ground truth(ICC=0.232),as well as between formulabased and automated measurements(ICC=0.186).In contrast,automated measurements showed no significant difference and excellent agreement with ground truth(ICC=0.915).Conclusion The Trans-UNet model enables accurate,efficient automated placental segmentation.Its volume measurements demonstrate high concordance with manual standards,supporting its potential as a foundational tool for automated placental volumetry in PAS diagnosis.
作者
黄竹媛
李楠楠
张玉恩
刘伟
闫锐
HUANG Zhu-yuan;LI Nan-nan;ZHANG Yu-en;LIU Wei;YAN Rui(Medical Image Centre,Northwest Women's and Children's Hospital,Xi'an 710061,Shaanxi Province,China;Department of Graduate Work,Xi'an Medical University,Xi'an 710021,Shaanxi Province,China;School of Computer Science&Technology,Xi'an University of Posts&Telecommunications,Xi'an 710121,Shaanxi Province,China)
出处
《中国CT和MRI杂志》
2026年第1期134-137,共4页
Chinese Journal of CT and MRI
基金
陕西省重点研发计划(2024SF-YBXM-239)。
关键词
深度学习
自动分割
胎盘体积
胎盘植入性疾病
Deep Learning
Automated Segmentation
Placental Volume
Placenta Accreta Spectrum Disorders
