A multi-path 2.5 dimensional convolutional neural network system for segmenting stroke lesions in brain MRI images
A multi-path 2.5 dimensional convolutional neural network system for segmenting stroke lesions in brain MRI images
复制标题
DOI：
10.1016/j.nicl.2019.102118
复制
发表时间：
2020-01-01
影响因子：
4.2
通讯作者：
Graves, William W.
中科院分区：
文献类型：
--
作者：
Xue, Yunzhe;Farhat, Fadi G.;Graves, William W.
从中风幸存者的磁共振成像（MRI）扫描中自动识别脑部病变将有助于患者诊断和治疗计划。通过消除让人类专家在每次脑部扫描中手动分割病变的费力步骤，它还将极大地促进大脑行为关系的研究。我们提出了一种多模态多路径卷积神经网络系统，用于自动中风病变分割。我们的系统有九个端到端 UNet，它们以二维 (2D) 切片作为输入，并使用三种不同的归一化检查所有三个平面。这九个总路径的输出被连接成一个 3D 体积，然后传递到 3D 卷积神经网络以输出最终的病变掩模。我们在来自三个来源的数据集上训练和测试了我们的方法：威斯康星医学院 (MCW)、凯斯勒基金会 (KF) 和公开的中风后病变解剖追踪 (ATLAS) 数据集。为了促进广泛的适用性，病变包括中风后亚急性（1至5周）和慢性（> 3个月）阶段，并且具有出血性和缺血性病因。获得交叉研究验证结果（具有独立的训练和验证数据集），以与之前基于朴素贝叶斯、随机森林和三个最近发布的卷积神经网络的方法进行比较。模型性能根据 Dice 系数进行量化，Dice 系数是模型识别的病变与人类专家识别的病变之间空间重叠的度量，其中 0 表示没有重叠，1 表示完全重叠。对 KF 和 MCW 图像进行训练以及对 ATLAS 图像进行测试，得出平均 Dice 系数为 0.54。这确实比之前最好的模型 UNet（0.47）要好。颠倒训练和测试数据集，KF 和 MCW 图像上的平均 Dice 为 0.47，而次优的 UNet 达到 0.45。将所有三个数据集结合起来，与以前的方法相比，当前系统还获得了可靠的更高的交叉验证精度。它还对许多现有方法难以识别的较小病变实现了高 Dice 值。总体而言，我们的系统比以前的自动中风病变分割方法有了明显的改进，使我们更接近人类专家的评估者间准确度水平。
Automatic identification of brain lesions from magnetic resonance imaging (MRI) scans of stroke survivors would be a useful aid in patient diagnosis and treatment planning. It would also greatly facilitate the study of brain-behavior relationships by eliminating the laborious step of having a human expert manually segment the lesion on each brain scan. We propose a multi-modal multi-path convolutional neural network system for automating stroke lesion segmentation. Our system has nine end-to-end UNets that take as input 2-dimensional (2D) slices and examines all three planes with three different normalizations. Outputs from these nine total paths are concatenated into a 3D volume that is then passed to a 3D convolutional neural network to output a final lesion mask. We trained and tested our method on datasets from three sources: Medical College of Wisconsin (MCW), Kessler Foundation (KF), and the publicly available Anatomical Tracings of Lesions After Stroke (ATLAS) dataset. To promote wide applicability, lesions were included from both subacute (1 to 5 weeks) and chronic (> 3 months) phases post stroke, and were of both hemorrhagic and ischemic etiology. Cross-study validation results (with independent training and validation datasets) were obtained to compare with previous methods based on naive Bayes, random forests, and three recently published convolutional neural networks. Model performance was quantified in terms of the Dice coefficient, a measure of spatial overlap between the model-identified lesion and the human expert-identified lesion, where 0 is no overlap and 1 is complete overlap. Training on the KF and MCW images and testing on the ATLAS images yielded a mean Dice coefficient of 0.54. This was reliably better than the next best previous model, UNet, at 0.47. Reversing the train and test datasets yields a mean Dice of 0.47 on KF and MCW images, whereas the next best UNet reaches 0.45. With all three datasets combined, the current system compared to previous methods also attained a reliably higher crossvalidation accuracy. It also achieved high Dice values for many smaller lesions that existing methods have difficulty identifying. Overall, our system is a clear improvement over previous methods for automating stroke lesion segmentation, bringing us an important step closer to the inter-rater accuracy level of human experts.