首页|人工智能技术在Stanford A型主动脉夹层CT血管成像图像分割中的应用

人工智能技术在Stanford A型主动脉夹层CT血管成像图像分割中的应用

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
目的 探讨Stanford B型主动脉夹层CT血管成像(CT angiography,CTA)人工智能分割软件的迁移学习在Stanford A型主动脉夹层诊断中的应用价值.方法 回顾性分析2018年12月-2021年7月阜外华中心血管病医院行手术治疗的41例Stanford A型主动脉夹层患者的临床资料.41例术前均行主动脉CTA,分别采用手工、Stanford B型主动脉夹层CTA人工智能分割软件及其迁移学习方式勾画主动脉,分割主动脉真腔和假腔掩膜.采用Dice相似系数分析迁移学习前后软件分割结果与医师勾画结果的相似度,比较迁移学习前后软件分割、医师勾画主动脉真腔和假腔掩膜的时间和诊断时间.结果 41例均成功完成图像收集、分割和标注勾画.迁移学习后软件分割主动脉真腔和假腔掩膜的结果与医师标记勾画结果的Dice相似系数(0.79±0.13,0.73±0.17)高于迁移学习前(0.72±0.11,0.61±0.28)(t=8.402,P＜0.001;t=3.148,P=0.003).迁移学习前、后软件分割1例患者真腔和假腔掩膜的时间均为(103±13)s.迁移学习后医师勾画1例患者真腔和假腔掩膜的时间[(150±25)min]和诊断时间[(3±1)min]均短于迁移学习前[(210±30)、(15±7)min](t=-50.310,P＜0.001;t=-12.366,P＜0.001).结论 Stanford B 型主动脉夹层 CTA图像人工智能分割软件迁移学习可提高Stanford A型主动脉夹层的诊断效率.
Artificial intelligence in CT angiography images segmentation of Stanford type A aortic dissection
Objective To investigate the application value of transfer learning based on artificial intelligence(AI)segmentation software for CT angiography(CTA)images of Stanford type B aortic dissection in the diagnosis of Stanford type A aortic dissection.Methods Forty-one patients with Stanford type A aortic dissection underwent surgical treatment in Fuwai Central China Cardiovascular Hospital from December 2018 to July 2021,and their clinical data were retrospectively analyzed.Before surgery,all patients received aortic CTA.The aorta was outlined manually or using transfer learning based on AI segmentation software for CTA images of Stanford type B aortic dissection,respectively,and the aortic true-and false-lumen masks were segmented.Dice similarity coefficient was used to analyze the similarity between the software segmentation results before and after transfer learning and the results outlined by radiologists,and the time for software segmentation and radiologists delineation and diagnosis was compared.Results The image collection,segmentation,annotation and delineation were successfully completed in 41 patients.The Dice similarity coefficients for the software segmentation of true-and false-lumen masks after transfer learning(0.79±0.13,0.73±0.17)were higher than those before transfer learning(0.72±0.11,0.61±0.28)(t=8.402,P＜0.001;t=3.148,P=0.003).The time for the software to segment the true-and false-lumen masks of one patient was(103±13)s both before and after transfer learning.The time for the radiologists to delineate the true-and false-lumen masks of one patient[(150±25)min]and the diagnosis time[(3±1)min]after transfer learning were shorter than those before transfer learning[(210±30),(15±7)min](t=-50.310,P＜0.001;t=-12.366,P＜0.001).Conclusion Transfer learning based on AI segmentation software for CT A images of Stanford type B aortic dissection can improve the diagnostic efficiency of Stanford type A aortic dissection.

aortic dissectionStanford type BStanford type Aartificial intelligenceCT angiographyimage segmentationtransfer learningDice similarity coefficient

谢瑞刚、潘玉坤、阚晓婧、葛英辉

展开 >

阜外华中心血管病医院放射科河南省心脏病影像医学重点实验室河南省人民医院心脏中心,河南郑州 451464

主动脉夹层 Stanford B型 Stanford A型 人工智能 CT血管成像 图像分割 迁移学习 Dice相似系数

河南省医学科技攻关计划联合共建项目河南省心脏病影像医学重点实验室

2018020453豫卫科教函[2021]44号

2024

10.13507/j.issn.1674-3474.2024.09.019

中华实用诊断与治疗杂志
中华预防医学会 河南省人民医院

中华实用诊断与治疗杂志

CSTPCD
影响因子：1.276
ISSN：1674-3474
年,卷(期)：2024.38(9)
  • 1