Liver shape analysis using partial least squares regression-based statistical shape model: application for understanding and staging of liver fibrosis
Liver shape analysis using partial least squares regression-based statistical shape model: application for understanding and staging of liver fibrosis
复制标题
DOI：
10.1007/s11548-019-02084-z
复制
发表时间：
2019-11-08
影响因子：
3
通讯作者：
Sato, Yoshinobu
中科院分区：
文献类型：
--
作者：
Soufi, Mazen;Otake, Yoshito;Sato, Yoshinobu
目的肝脏形态变化被认为是肝纤维化的可行指标。然而，目前的统计形状模型（SSM）的基础上，主成分分析代表总的形状变化，而不考虑与纤维化阶段。因此，我们的目标是应用统计形状建模方法，使用偏最小二乘回归（PLSR），明确使用阶段作为监督信息，了解与阶段相关的形状变化，以及预测它在对比增强MR图像。方法对51例肝纤维化患者行MRI增强扫描，其中F0/1期18例，F2期15例，F3期7例，F4期11例。从图像中手动分割肝脏。使用PLSR构建SSM，通过PLSR导出与参考病理性纤维化阶段明确相关的形状变化模式（评分）。使用基于PLSR分数的支持向量机（SVM）预测该阶段。使用受试者工作特征曲线下面积（AUC）评估性能。结果除了常见的左叶增大、右叶缩小等形态变化外，模型还表现了尾状叶、右叶后部增大、右叶前部缩小等细节变化。这些变化定性上与临床研究中报告的局部体积变化一致。F0/1与F2-4（显著纤维化）、F0-2与F3-4和F0-3与F4（肝硬化）分类的准确度（AUC）分别为0.90 +/- 0.03、0.80 +/- 0.05和0.82 +/- 0.05。结论所提出的方法提供了一个明确的代表性，以及众所周知的详细形状变化与肝纤维化阶段。因此，基于PLSR的SSM的应用对于理解与肝纤维化阶段相关的形状变化并预测它是可行的。
Purpose Liver shape variations have been considered as feasible indicators of liver fibrosis. However, current statistical shape models (SSM) based on principal component analysis represent gross shape variations without considering the association with the fibrosis stage. Therefore, we aimed at the application of a statistical shape modelling approach using partial least squares regression (PLSR), which explicitly uses the stage as supervised information, for understanding the shape variations associated with the stage as well as predicting it in contrast-enhanced MR images. Methods Contrast-enhanced MR images of 51 patients with fibrosis stages F0/1 (n = 18), F2 (n = 15), F3 (n = 7) and F4 (n = 11) were used. The livers were manually segmented from the images. An SSM was constructed using PLSR, by which shape variation modes (scores) that were explicitly associated with the reference pathological fibrosis stage were derived. The stage was predicted using a support vector machine (SVM) based on the PLSR scores. The performance was assessed using the area under receiver operating characteristic curve (AUC). Results In addition to commonly known shape variations, such as enlargement of left lobe and shrinkage of right lobe, our model represented detailed variations, such as enlargement of caudate lobe and the posterior part of right lobe, and shrinkage in the anterior part of right lobe. These variations qualitatively agreed with localized volumetric variations reported in clinical studies. The accuracy (AUC) at classifications F0/1 versus F2-4 (significant fibrosis), F0-2 versus F3-4 and F0-3 versus F4 (cirrhosis) were 0.90 +/- 0.03, 0.80 +/- 0.05 and 0.82 +/- 0.05, respectively. Conclusions The proposed approach offered an explicit representation of commonly known as well as detailed shape variations associated with liver fibrosis stage. Thus, the application of PLSR-based SSM is feasible for understanding the shape variations associated with the liver fibrosis stage and predicting it.