Skip to main content
QUICK REVIEW

[论文解读] Automatic Spine Segmentation using Convolutional Neural Network via Redundant Generation of Class Labels for 3D Spine Modeling

Malinda Vania, Dawit Mureja|arXiv (Cornell University)|Nov 29, 2017
Medical Imaging and Analysis参考文献 19被引用 8
一句话总结

本文提出了一种基于CNN的新方法,通过使用冗余类别标签作为软约束,实现从CT扫描中自动进行3D脊柱分割。通过将类别冗余整合到混合CNN-FCN架构中,该方法在真实患者数据上显著优于经典CNN和U-Net等基线模型,实现了高精度(Dice:94%)和低误差率(MSD:0.16 mm,GCE:0.02)。

ABSTRACT

There has been a significant increase from 2010 to 2016 in the number of people suffering from spine problems. The automatic image segmentation of the spine obtained from a computed tomography (CT) image is important for diagnosing spine conditions and for performing surgery with computer-assisted surgery systems. The spine has a complex anatomy that consists of 33 vertebrae, 23 intervertebral disks, the spinal cord, and connecting ribs. As a result, the spinal surgeon is faced with the challenge of needing a robust algorithm to segment and create a model of the spine. In this study, we developed an automatic segmentation method to segment the spine, and we compared our segmentation results with reference segmentations obtained by experts. We developed a fully automatic approach for spine segmentation from CT based on a hybrid method. This method combines the convolutional neural network (CNN) and fully convolutional network (FCN), and utilizes class redundancy as a soft constraint to greatly improve the segmentation results. The proposed method was found to significantly enhance the accuracy of the segmentation results and the system processing time. Our comparison was based on 12 measurements: the Dice coefficient (94%), Jaccard index (93%), volumetric similarity (96%), sensitivity (97%), specificity (99%), precision (over segmentation; 8.3 and under segmentation 2.6), accuracy (99%), Matthews correlation coefficient (0.93), mean surface distance (0.16 mm), Hausdorff distance (7.4 mm), and global consistency error (0.02). We experimented with CT images from 32 patients, and the experimental results demonstrated the efficiency of the proposed method.

研究动机与目标

  • 解决由于解剖变异、噪声和低对比度结构导致的复杂3D CT图像中脊柱自动分割的挑战。
  • 减少对手动干预的依赖,并提高对33个椎体、23个椎间盘及相关组织分割的鲁棒性。
  • 通过在深度学习框架内引入类别冗余作为软约束,提升分割性能。
  • 通过完全自动化的端到端方法,在多种脊柱结构中实现高精度和一致性。

提出的方法

  • 采用结合卷积神经网络(CNN)和全卷积网络(FCN)的混合深度学习架构,用于3D脊柱分割。
  • 生成冗余类别标签以编码结构先验知识,利用类别冗余作为软约束来正则化预测并提升泛化能力。
  • 在32例患者CT扫描的1,388,800个脊柱切片上进行训练,采用有监督学习方法,使用专家标注的真实标签作为监督信号。
  • 应用基于距离的指标,包括平均表面距离(MSD)、豪斯多夫距离(HD)和全局一致性误差(GCE),进行定量评估。
  • 使用Dice系数、Jaccard指数和马修斯相关系数(MCC)评估预测分割与真实标签之间的重叠程度和相关性。
  • 在CPU上进行训练(完整数据集耗时13小时),并实现每例CT体积约3分钟的推理时间。

实验结果

研究问题

  • RQ1在无需手动调参的情况下,冗余类别标签是否能提升复杂3D脊柱结构中的分割精度?
  • RQ2所提出的CNN-FCN混合架构结合类别冗余与标准CNN、U-Net及水平集方法相比,在分割指标上表现如何?
  • RQ3类别冗余的软约束在多大程度上减少了脊柱CT图像中的过度分割和欠分割错误?
  • RQ4该方法是否能在多种解剖变异下实现高特异性与高敏感性,同时保持低表面距离和全局一致性误差?

主要发现

  • 所提方法的Dice系数达到94%,表明预测分割与真实标签之间具有强重叠。
  • Jaccard指数达到93%,进一步证实了分割结果在结构变异下的高度相似性与鲁棒性。
  • 平均表面距离(MSD)为0.16 mm,表明分割表面与专家标注在几何对齐上具有高精度。
  • 豪斯多夫距离为7.4 mm,表明分割表面与真实标签之间的最大误差极小。
  • 全局一致性误差(GCE)为0.02,表明在体素层面的平均误差极低,分割结果高度可靠。
  • 敏感性(97%)和特异性(99%)均较高,证实了对脊柱结构的有效检测,且假阴性和假阳性极少。

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。