Quantitative lung morphology: semi-automated measurement of mean linear intercept
Quantitative lung morphology: semi-automated measurement of mean linear intercept
复制标题
DOI：
10.1186/s12890-019-0915-6
复制
发表时间：
2019-11-09
影响因子：
3.1
通讯作者：
Nolan, Anna
中科院分区：
文献类型：
--
作者：
Crowley, George;Kwon, Sophia;Nolan, Anna
背景定量的形态变化对于我们理解肺的病理生理学是至关重要的。平均线性截距(MLI)测量在评估临床相关病理中非常重要，例如肺气肿。然而，定性方法容易产生误差和偏差，而平均线性截距(MLI)等定量方法人工费时。此外，一种完全自动化、可靠的评估方法是非常重要的，而且是资源密集型的。方法我们提出了一种半自动的方法来量化MLI，该方法不需要专门的计算机知识，并且使用免费的、开源的图像处理器(斐济)。我们用计算机生成的理想化数据集测试了该方法，得出了MLI使用指南，并成功地将该方法应用于颗粒物(PM)暴露的小鼠模型。分析了随机放置的等半径圆的场。通过受试者-操作者特征(ROC)-曲线下面积(AUC)分析，找出基于MLI评估的最佳弦数。组内相关系数(ICC)测量信度。结果我们表现出很高的准确性(AUC(ROC)>；0.8，MLIActual&Gt；63.83像素)和极好的可靠性(ICC=0.9998，p<；0.0001)。我们提供了一个指南，以优化和弦的数量样本基于MLI。处理时间为0.03张S/张。我们发现，与PBS暴露的对照组相比，PM暴露的小鼠MLI升高。我们还提供了使用过的宏，并提供了一个免费的ImageJ插件供学术研究使用。结论我们的半自动方法是可靠的，与全自动方法一样快，并且使用了免费的开源软件。此外，我们量化了每个肺野应该测量的最佳弦数。
Background Quantifying morphologic changes is critical to our understanding of the pathophysiology of the lung. Mean linear intercept (MLI) measures are important in the assessment of clinically relevant pathology, such as emphysema. However, qualitative measures are prone to error and bias, while quantitative methods such as mean linear intercept (MLI) are manually time consuming. Furthermore, a fully automated, reliable method of assessment is nontrivial and resource-intensive. Methods We propose a semi-automated method to quantify MLI that does not require specialized computer knowledge and uses a free, open-source image-processor (Fiji). We tested the method with a computer-generated, idealized dataset, derived an MLI usage guide, and successfully applied this method to a murine model of particulate matter (PM) exposure. Fields of randomly placed, uniform-radius circles were analyzed. Optimal numbers of chords to assess based on MLI were found via receiver-operator-characteristic (ROC)-area under the curve (AUC) analysis. Intraclass correlation coefficient (ICC) measured reliability. Results We demonstrate high accuracy (AUC(ROC) > 0.8 for MLIactual > 63.83 pixels) and excellent reliability (ICC = 0.9998, p < 0.0001). We provide a guide to optimize the number of chords to sample based on MLI. Processing time was 0.03 s/image. We showed elevated MLI in PM-exposed mice compared to PBS-exposed controls. We have also provided the macros that were used and have made an ImageJ plugin available free for academic research use at . Conclusions Our semi-automated method is reliable, equally fast as fully automated methods, and uses free, open-source software. Additionally, we quantified the optimal number of chords that should be measured per lung field.