Fully automatic, multiorgan segmentation in normal whole body magnetic resonance imaging (MRI), using classification forests (CFs), convolutional neural networks (CNNs), and a multi-atlas (MA) approach
Fully automatic, multiorgan segmentation in normal whole body magnetic resonance imaging (MRI), using classification forests (CFs), convolutional neural networks (CNNs), and a multi-atlas (MA) approach
复制标题
DOI：
10.1002/mp.12492
复制
发表时间：
2017-10-01
期刊：
影响因子：
3.8
通讯作者：
Rockall, Andrea G.
中科院分区：
文献类型：
--
作者：
Lavdas, Ioannis;Glocker, Ben;Rockall, Andrea G.
目的：作为实现肿瘤全身磁共振成像(MRI)病变自动检测方法的一部分，我们开发、评估和比较了三种全自动多器官分割算法。方法：第一种算法基于分类森林(CFS)，第二种算法基于三维卷积神经网络(CNN)，第三种算法基于多图谱(MA)方法。我们检查了51名健康志愿者的数据，他们在1.5T下使用标准化的多参数全身磁共振扫描方案进行前瞻性扫描。研究得到了当地伦理委员会的批准，并获得了参与者的书面同意。MRI数据被用作算法的输入数据，而训练则基于临床MRI专家对感兴趣的解剖结构的手动注释。对34名无伪影的受试者进行了五次交叉验证实验。我们报告了三个重叠和三个表面距离度量来评估自动分割和手动分割之间的一致性，即骰子相似系数(DSC)、召回率(RE)、精度(PR)、平均表面距离(ASD)、均方根表面距离(RMSSD)和Hausdorff距离(HD)。在每个器官的基础上，使用方差分析来比较三种算法和DSC之间的合并标签度量。当使用不同的成像组合作为训练的输入时，使用Mann-Whitney U检验来比较CFS和CNN之间以及每个器官的DSC之间的合并指标。结果：所有三种算法都产生了健壮的分段器，并且使用相对较少的数据集进行了有效的训练，这在临床环境中是一个重要的考虑因素。CFS：DSC=0.70+/-0.18，RE=0.73+/-0.18，PR=0.71+/-0.14，CNNS：DSC=0.81+/-0.13，RE=0.83+/-0.14，PR=0.82+/-0.10，MA：DSC=0.71+/-0.22，RE=0.70+/-0.34，PR=0.77+/-0.15。所有节段结构的平均表面距离为：CFS：ASD=13.5+/-11.3 mm，RMSSD=34.6+/-37.6 mm，HD=185.7+/-194.0 mm，CNN；ASD=5.48±/-4.84 mm，RMSSD=17.0±/-13.3 mm，HD=199.0±/-101.2 mm；当使用所有成像组合(T2w+T1w+DWI)作为输入时，CFS的共用性能得到改善，而CNN的性能下降，但这两种情况下都没有显著差异。以T2w图像为输入的CNN在除膀胱外的所有解剖标记物上的分割效果均明显优于CFS。结论：开发了三种先进的算法，并用于全身MRI中主要器官和骨骼的自动分割，与临床MRI专家的手动分割结果吻合较好。当使用T2w卷作为输入时，CNN表现良好。使用多模式MRI数据作为CNN的输入并不能提高分割性能。(C)2017年美国医学物理学家协会
Purpose: As part of a program to implement automatic lesion detection methods for whole body magnetic resonance imaging (MRI) in oncology, we have developed, evaluated, and compared three algorithms for fully automatic, multiorgan segmentation in healthy volunteers.Methods: The first algorithm is based on classification forests (CFs), the second is based on 3D convolutional neural networks (CNNs) and the third algorithm is based on a multi-atlas (MA) approach. We examined data from 51 healthy volunteers, scanned prospectively with a standardized, multiparametric whole body MRI protocol at 1.5 T. The study was approved by the local ethics committee and written consent was obtained from the participants. MRI data were used as input data to the algorithms, while training was based on manual annotation of the anatomies of interest by clinical MRI experts. Fivefold cross-validation experiments were run on 34 artifact-free subjects. We report three overlap and three surface distance metrics to evaluate the agreement between the automatic and manual segmentations, namely the dice similarity coefficient (DSC), recall (RE), precision (PR), average surface distance (ASD), root-mean-square surface distance (RMSSD), and Hausdorff distance (HD). Analysis of variances was used to compare pooled label metrics between the three algorithms and the DSC on a per-organ' basis. A Mann-Whitney U test was used to compare the pooled metrics between CFs and CNNs and the DSC on a per-organ' basis, when using different imaging combinations as input for training.Results: All three algorithms resulted in robust segmenters that were effectively trained using a relatively small number of datasets, an important consideration in the clinical setting. Mean overlap metrics for all the segmented structures were: CFs: DSC = 0.70 0.18, RE = 0.73 +/- 0.18, PR = 0.71 +/- 0.14, CNNs: DSC = 0.81 +/- 0.13, RE = 0.83 +/- 0.14, PR = 0.82 +/- 0.10, MA: DSC = 0.71 +/- 0.22, RE = 0.70 +/- 0.34, PR = 0.77 +/- 0.15. Mean surface distance metrics for all the segmented structures were: CFs: ASD = 13.5 +/- 11.3 mm, RMSSD = 34.6 +/- 37.6 mm and HD = 185.7 +/- 194.0 mm, CNNs; ASD = 5.48 +/- 4.84 mm, RMSSD = 17.0 +/- 13.3 mm and HD = 199.0 +/- 101.2 mm, MA: ASD = 4.22 +/- 2.42 mm, RMSSD = 6.13 +/- 2.55 mm, and HD = 38.9 +/- 28.9 mm. The pooled performance of CFs improved when all imaging combinations (T2w + T1w + DWI) were used as input, while the performance of CNNs deteriorated, but in neither case, significantly. CNNs with T2w images as input, performed significantly better than CFs with all imaging combinations as input for all anatomical labels, except for the bladder.Conclusions: Three state-of-the-art algorithms were developed and used to automatically segment major organs and bones in whole body MRI; good agreement to manual segmentations performed by clinical MRI experts was observed. CNNs perform favorably, when using T2w volumes as input. Using multimodal MRI data as input to CNNs did not improve the segmentation performance. (C) 2017 American Association of Physicists in Medicine