Yunker Lauren, Harwig Megan Cleland, Kriegel Alison J
Department of Physiology, Medical College of Wisconsin, Milwaukee, Wisconsin, United States.
Department of Physiology, Medical College of Georgia, Augusta University, Augusta, Georgia, United States.
Am J Physiol Renal Physiol. 2025 Feb 1;328(2):F230-F238. doi: 10.1152/ajprenal.00252.2024. Epub 2024 Dec 24.
The presence of tubular casts within the kidney serves as an important feature when assessing the degree of renal injury. Quantification of renal tubular casts has been historically difficult due to varying cast morphologies, protein composition, and stain uptake properties, even within the same kidney. Color thresholding remains one of the most common methods of quantification in the laboratory when assessing the percentage of renal casting; however, this method is unable to account for tubule casts stained a variety of colors. We have developed a novel method of automated cast quantification using the machine learning pixel classification tool within QuPath, an open-source software designed for digital pathology. We demonstrated the usability of this method in male and female Dahl salt-sensitive rats fed either low or high salt for 2 wk and male Sprague-Dawley rats treated with podotoxin puromycin aminonucleoside (PAN). Briefly, the pixel classifier was trained to identify kidney tissue, various cast color types, and slide backgrounds. Following the development of the pixel classifier, we applied it to the sample population and compared the results with those of other methods of cast quantification, including color thresholding and manual quantification. We found that the automated pixel classifier designed in QuPath was able to comprehensively quantify metachromatic tubular casts compared with color thresholding. This novel method of cast quantification provides researchers with the ability to reliably automate cast quantification that is both comprehensive and efficient. We developed a method of automated renal tubule cast quantification using a machine learning-based pixel classifier within QuPath, an open-source image analysis software. The advantages of this approach are demonstrated by rigorous comparison of quantification methods on a set of Masson's trichrome-stained kidney sections from high- and low-salt fed salt-sensitive Dahl rats. Researchers are provided with step-by-step instructions for creating and training a pixel classifier in QuPath for application to image analysis.
在评估肾损伤程度时，肾内出现肾小管管型是一个重要特征。由于管型形态、蛋白质组成和染色摄取特性各不相同，即使在同一个肾脏内，对肾小管管型进行定量分析历来都很困难。在实验室评估肾铸型百分比时，颜色阈值法仍然是最常用的定量方法之一；然而，这种方法无法对染有多种颜色的肾小管管型进行计数。我们开发了一种新的自动管型定量方法，使用QuPath中的机器学习像素分类工具，QuPath是一款专为数字病理学设计的开源软件。我们在喂食低或高盐2周的雄性和雌性Dahl盐敏感大鼠以及用嘌呤霉素氨基核苷（PAN）处理的雄性Sprague-Dawley大鼠中证明了该方法的可用性。简而言之，像素分类器经过训练，可识别肾组织、各种管型颜色类型和玻片背景。在开发像素分类器之后，我们将其应用于样本群体，并将结果与其他管型定量方法（包括颜色阈值法和手动定量法）的结果进行比较。我们发现，与颜色阈值法相比，QuPath中设计的自动像素分类器能够全面地对异染性肾小管管型进行定量。这种新的管型定量方法为研究人员提供了可靠地自动进行管型定量的能力，既全面又高效。我们使用开源图像分析软件QuPath中基于机器学习的像素分类器开发了一种自动肾小管管型定量方法。通过对一组来自高盐和低盐喂养的盐敏感Dahl大鼠的Masson三色染色肾脏切片的定量方法进行严格比较，证明了这种方法的优势。为研究人员提供了在QuPath中创建和训练像素分类器以应用于图像分析的分步说明。
