三维心脏运动估计网络的实现与验证
Implementation and Validation of a Three-dimensional Cardiac Motion Estimation Network.
作者信息
Morales Manuel A, Izquierdo-Garcia David, Aganj Iman, Kalpathy-Cramer Jayashree, Rosen Bruce R, Catana Ciprian
机构信息
Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, 149 13th St, Charlestown, MA 02129 (M.A.M., D.I.G., I.A., J.K.C., B.R.R., C.C.); Harvard-MIT Division of Health Sciences and Technology (M.A.M.) and Computer Science and Artificial Intelligence Laboratory (I.A.), Massachusetts Institute of Technology, Cambridge, Mass.
出版信息
Radiol Artif Intell. 2019 Jul 17;1(4):e180080. doi: 10.1148/ryai.2019180080.
PURPOSE
To describe an unsupervised three-dimensional cardiac motion estimation network (CarMEN) for deformable motion estimation from two-dimensional cine MR images.
MATERIALS AND METHODS
A function was implemented using CarMEN, a convolutional neural network that takes two three-dimensional input volumes and outputs a motion field. A smoothness constraint was imposed on the field by regularizing the Frobenius norm of its Jacobian matrix. CarMEN was trained and tested with data from 150 cardiac patients who underwent MRI examinations and was validated on synthetic ( = 100) and pediatric ( = 33) datasets. CarMEN was compared to five state-of-the-art nonrigid body registration methods by using several performance metrics, including Dice similarity coefficient (DSC) and end-point error.
RESULTS
On the synthetic dataset, CarMEN achieved a median DSC of 0.85, which was higher than all five methods (minimum-maximum median [or MMM], 0.67-0.84; < .001), and a median end-point error of 1.7, which was lower than (MMM, 2.1-2.7; < .001) or similar to (MMM, 1.6-1.7; > .05) all other techniques. On the real datasets, CarMEN achieved a median DSC of 0.73 for Automated Cardiac Diagnosis Challenge data, which was higher than (MMM, 0.33; < .0001) or similar to (MMM, 0.72-0.75; > .05) all other methods, and a median DSC of 0.77 for pediatric data, which was higher than (MMM, 0.71-0.76; < .0001) or similar to (MMM, 0.77-0.78; > .05) all other methods. All values were derived from pairwise testing. For all other metrics, CarMEN achieved better accuracy on all datasets than all other techniques except for one, which had the worst motion estimation accuracy.
CONCLUSION
The proposed deep learning-based approach for three-dimensional cardiac motion estimation allowed the derivation of a motion model that balances motion characterization and image registration accuracy and achieved motion estimation accuracy comparable to or better than that of several state-of-the-art image registration algorithms.© RSNA, 2019
目的
描述一种用于从二维电影磁共振图像进行可变形运动估计的无监督三维心脏运动估计网络（CarMEN）。
材料与方法
使用CarMEN实现了一个函数，这是一个卷积神经网络，它接受两个三维输入体积并输出一个运动场。通过对其雅可比矩阵的弗罗贝尼乌斯范数进行正则化，对该场施加了平滑约束。使用来自150名接受MRI检查的心脏病患者的数据对CarMEN进行训练和测试，并在合成（n = 100）和儿科（n = 33）数据集上进行验证。通过使用包括骰子相似系数（DSC）和端点误差在内的几个性能指标，将CarMEN与五种最先进的非刚体配准方法进行比较。
结果
在合成数据集上，CarMEN的DSC中位数为0.85，高于所有五种方法（最小 - 最大中位数[或MMM]，0.67 - 0.84；P <.001），端点误差中位数为1.7，低于（MMM，2.1 - 2.7；P <.001）或与所有其他技术相似（MMM，1.6 - 1.7；P>.05）。在真实数据集上，对于自动心脏诊断挑战数据，CarMEN的DSC中位数为0.73，高于（MMM，0.33；P <.0（此处原文有误，推测是P <.0001））或与所有其他方法相似（MMM，0.72 - 0.75；P>.05），对于儿科数据，DSC中位数为0.77，高于（MMM，0.71 - 0.76；P <.0001）或与所有其他方法相似（MMM，0.77 - 0.78；P>.05）。所有P值均来自成对测试。对于所有其他指标，除了一种运动估计精度最差的方法外，CarMEN在所有数据集上的准确性均优于所有其他技术。
结论
所提出的基于深度学习的三维心脏运动估计方法允许推导一个平衡运动表征和图像配准精度的运动模型，并实现了与几种最先进的图像配准算法相当或更好的运动估计精度。©RSNA，2019
Zotero 插件