[论文解读] Automatic Cerebral Vessel Extraction in TOF-MRA Using Deep Learning

Valentina Vos, Kimberley M. Timmins+4 more|arXiv (Cornell University)|Jan 22, 2021

Advanced MRI Techniques and Applications参考文献 15被引用 7

一句话总结

本研究提出一种基于增强64×64×64体素块的3D U-Net深度学习模型，用于自动分割时间飞越磁共振血管成像（TOF-MRA）扫描中的脑血管。通过使用高斯模糊、旋转和翻转进行数据增强，该方法在测试数据上实现了高达0.83的Dice相似系数（DSC），表现最优。

ABSTRACT

Deep learning approaches may help radiologists in the early diagnosis and timely treatment of cerebrovascular diseases. Accurate cerebral vessel segmentation of Time-of-Flight Magnetic Resonance Angiographs (TOF-MRAs) is an essential step in this process. This study investigates deep learning approaches for automatic, fast and accurate cerebrovascular segmentation for TOF-MRAs. The performance of several data augmentation and selection methods for training a 2D and 3D U-Net for vessel segmentation was investigated in five experiments: a) without augmentation, b) Gaussian blur, c) rotation and flipping, d) Gaussian blur, rotation and flipping and e) different input patch sizes. All experiments were performed by patch-training both a 2D and 3D U-Net and predicted on a test set of MRAs. Ground truth was manually defined using an interactive threshold and region growing method. The performance was evaluated using the Dice Similarity Coefficient (DSC), Modified Hausdorff Distance and Volumetric Similarity, between the predicted images and the interactively defined ground truth. The segmentation performance of all trained networks on the test set was found to be good, with DSC scores ranging from 0.72 to 0.83. Both the 2D and 3D U-Net had the best segmentation performance with Gaussian blur, rotation and flipping compared to other experiments without augmentation or only one of those augmentation techniques. Additionally, training on larger patches or slices gave optimal segmentation results. In conclusion, vessel segmentation can be optimally performed on TOF-MRAs using a trained 3D U-Net on larger patches, where data augmentation including Gaussian blur, rotation and flipping was performed on the training data.

研究动机与目标

• 开发一种自动、快速且准确的脑血管分割方法，用于时间飞越磁共振血管成像（TOF-MRA）扫描。
• 评估不同数据增强和块选择策略对2D和3D U-Net架构分割性能的影响。
• 确定用于TOF-MRA脑血管分割的最优网络架构（2D与3D）及训练配置。
• 建立一种稳健的深度学习分割流程，以减少对手动分割的依赖。
• 通过实现快速、可重复的3D血管建模，提升临床工作流程效率，助力脑血管疾病早期诊断。

提出的方法

• 在TOF-MRA扫描中随机选取并增强的图像块上，分别训练2D和3D全卷积U-Net架构。
• 使用64×64个体素（2D）和16×16×16个体素（3D）的块大小，并对血管区域（80%）与非血管区域（20%）进行平衡采样。
• 应用高斯模糊、旋转和翻转等数据增强技术，以提升模型的泛化能力和鲁棒性。
• 作为预处理步骤，执行N4偏置场校正和Z得分标准化，以提高图像质量和一致性。
• 通过MevisLab中的交互式阈值法和区域生长法生成真实标签，使用最大强度的95%至99%之间的强度阈值。
• 在保留的测试集上，使用Dice相似系数（DSC）、修正马氏距离（MHD）和体积相似性评估模型性能。

Figure 1 : Example slice in the transverse plane of a TOF-MRA.

实验结果

研究问题

• RQ1 数据增强（高斯模糊、旋转、翻转）如何影响2D和3D U-Net模型在TOF-MRA脑血管分割中的性能？
• RQ2 在分割准确性和泛化能力方面，2D和3D U-Net模型的最优块尺寸是什么？
• RQ3 2D和3D U-Net架构在TOF-MRA脑血管分割中的性能表现如何比较？
• RQ4 基于深度学习的方法能否实现与人工分割相当的高分割精度，同时降低观察者间和观察者内差异性？
• RQ5 与较小块相比，使用更大块（如64×64×64个体素）进行训练是否能提升分割性能？

主要发现

• 在64×64×64个体素块上，结合高斯模糊、旋转和翻转增强的3D U-Net模型，实现了最高的Dice相似系数（DSC）0.83。
• 三种增强技术（高斯模糊、旋转、翻转）的组合相比单一或无增强方法，显著提升了分割性能。
• 较大的块尺寸（64×64×64个体素）相比小块尺寸带来了更优的分割结果，可能是因为其能更好地表征小血管的上下文信息。
• 2D和3D U-Net模型均表现出良好性能，所有实验的DSC得分在0.72至0.83之间。
• 修正马氏距离（MHD）相对较高，表明在准确捕捉小血管或断裂血管方面仍存在挑战，可能源于交互式真实标签生成方法的局限性。
• 模型推理时间在每幅图像数秒量级，支持其在临床中实现快速血管分割的可行性。

Figure 2 : Boxplots of the vessel segmentation results obtained by training a 2D U-Net. The evaluation was performed by five experiments: A) without augmentation; B) augmented training data with Gaussian blurring; C) augmented training data with rotation and flipping; D) augmented training data with