摘要
目的对比基于深度学习的脑影像量化分析软件QBrain与作为传统算法代表的FreeSurfer自动分割软件测量海马体积的效果。方法自健康中国成人头颅高分辨MRI数据库(Chinese 1000)中随机提取100名志愿者数据,以手动测量结果为标准,对比FreeSurfer、QBrain软件测量海马体积的差异。采用Pearson相关分析和组内相关系数(ICC)进行相关性分析,以Bland-Altman分析法行一致性分析;以戴斯系数(DSC)评估2种自动测量方法与手动测量值相似度。结果手动测量及FreeSurfer、QBrain软件测量结果均显示右侧海马体积测值大于左侧(P均<0.05)。FreeSurfer右侧海马测值明显小于手动测值(P<0.05);左侧测值大于手动测值,但差异无统计学意义(P>0.05)。QBrain右侧海马值明显小于手动测值(P<0.05);左侧测值小于手动测值,但差异无统计学意义(P>0.05)。2种自动方法海马体积测值均与手动测值显著相关(ICC>0.75,r>0.75,P<0.05)。Bland-Altman散点图显示2种自动测量方法与手动测量结果的一致性均较好。FreeSurfer及QBrain海马体积测值与手动测值的DSC分别为0.73和0.05(P均<0.05)。结论FreeSurfer与QBrain均可准确自动测量海马体积;QBrain所测海马体积与手动测值的相关性更佳,且耗时较少。
Objective To compare the measurement effects of automatic segmentation software for hippocampus segmentation tasks between QBrain as a quantitative analysis software for brain images based on deep learning and FreeSurfer as a traditional algorithm representing automatic segmentation software.Methods Data of 100 volunteers were randomly selected from Healthy Chinese adult cranial MRI high-resolution database(Chinese 1000).FreeSurfer,QBrain and manual measurement results of hippocampal volume were compared.Pearson correlation analysis and intra-class correlation coefficient(ICC)were performed for correlation analysis,and Bland-Altman analysis was used for consistency analysis.The similarity between FreeSurfer,QBrain measuremens,also between manual measurements were evaluated with Dice similarity coefficient(DSC).Results The measurement values of the right side hippocampus were larger than those of the left side(all P<0.05).The FreeSurfer measured values of the right hippocampus were significantly smaller than manually measured values(P<0.05),of the left hippocampus were larger than manually measured values,but there was no statistical difference(P>0.05).The QBrain measured right values were significantly smaller than manually measured values(P<0.05),the QBrain measured left values were smaller than manually measured values,but the difference was not statistically significant(P>0.05).FreeSurfer and QBrain measured values were all significantly correlated with manually measured values of both left and right hippocampal volume(ICC>0.75,r>0.75,P<0.05).Bland-Altman scatter plot showed that both FreeSurfer and QBrain measurements were consistent with manual measurements.The DSC values of FreeSurfer and QBrain and manual measurement of hippocampal volume was 0.73 and 0.05,respectively(both P<0.05).Conclusion Both FreeSurfer and QBrain could measure hippocampal volume accurately and automatically.The correlation between QBrain and manually measured values of the hippocampal volume was better than that of FreeSurfer,and the measurement time of QBrain was shorter.
作者
王颖
钱天翼
苏壮志
张泽寅
李坤成
WANG Ying;QIAN Tianyi;SU Zhuangzhi;ZHANG Zeyin;LI Kuncheng(Department of Radiology and Nuclear Medicine,Xuanwu Hospital,Capital Medical University,Beijing 100053,China;Department of Radiology,Beijing Jiangong Hospital,Beijing 100054,China;Beijing Quantmind Itd,Beijing 100037,China;Department of Radiology,the Central Hospital of Beijing Prison,Beijing 100054,China;Beijing Key Laboratory of Magnetic Resonance Imaging and Brain Informatics,Beijing 100053,China)
出处
《中国医学影像技术》
CSCD
北大核心
2021年第2期174-178,共5页
Chinese Journal of Medical Imaging Technology
基金
国家重点研发计划(2020YFC2007302)
北京市科技计划(Z171100000117001)
北京市优秀人才青年拔尖团队项目(2016000021223TD07)
磁共振成像脑信息学北京重点实验室专项基金资助。
关键词
海马
体积测量
磁共振成像
hippocampus
volume measurement
magnetic resonance imaging