School of Medicine, Northwest University, Xi'an, Shaanxi, China.
Faculty of Medicine, School of Biomedical Sciences, The Chinese University of Hong Kong, Shatin, Hong Kong SAR, China.
PLoS One. 2023 Oct 5;18(10):e0286626. doi: 10.1371/journal.pone.0286626. eCollection 2023.
Quantification of the histological staining images gives important insights in biomedical research. In wet lab, it is common to have some stains off the target to become unwanted noisy stains during the generation of histological staining images. The current tools designed for quantification of histological staining images do not consider such situations; instead, the stained region is identified based on assumptions that the background is pure and clean. The goal of this study is to develop a light software named Staining Quantification (SQ) tool which could handle the image quantification job with features for removing a large amount of unwanted stains blended or overlaid with Region of Interest (ROI) in complex scenarios. The core algorithm was based on the method of higher order statistics transformation, and local density filtering. Compared with two state-of-art thresholding methods (i.e. Otsu's method and Triclass thresholding method), the SQ tool outperformed in situations such as (1) images with weak positive signals and experimental caused dirty stains; (2) images with experimental counterstaining by multiple colors; (3) complicated histological structure of target tissues. The algorithm was developed in R4.0.2 with over a thousand in-house histological images containing Alizarin Red (AR) and Von Kossa (VK) staining, and was validated using external images. For the measurements of area and intensity in total and stained region, the average mean of difference in percentage between SQ and ImageJ were all less than 0.05. Using this as a criterion of successful image recognition, the success rate for all measurements in AR, VK and external validation batch were above 0.8. The test of Pearson's coefficient, difference between SQ and ImageJ, and difference of proportions between SQ and ImageJ were all significant at level of 0.05. Our results indicated that the SQ tool is well established for automatic histological staining image quantification.
组织学染色图像的定量分析为生物医学研究提供了重要的见解。在湿实验室中，在生成组织学染色图像的过程中，常见的情况是一些目标以外的染色会变成不需要的噪点染色。目前设计用于组织学染色图像定量分析的工具没有考虑到这种情况；相反，染色区域是根据背景是纯净的假设来识别的。本研究的目的是开发一种名为染色定量（SQ）的轻量级软件工具，该工具可以处理具有大量去除与感兴趣区域（ROI）混合或叠加的不需要的染色的图像定量工作，适用于复杂情况。核心算法基于高阶统计变换方法和局部密度滤波方法。与两种最先进的阈值方法（即 Otsu 方法和三分类阈值方法）相比，SQ 工具在以下情况下表现更好：（1）图像具有较弱的阳性信号和实验引起的脏染色；（2）图像用多种颜色进行实验复染；（3）目标组织的复杂组织学结构。该算法是在 R4.0.2 中开发的，使用了超过一千张包含茜素红（AR）和 von Kossa（VK）染色的内部组织学图像，并使用外部图像进行了验证。对于总面积和染色区域的面积和强度测量，SQ 和 ImageJ 之间差异的平均百分比平均值均小于 0.05。将此作为图像识别成功的标准，AR、VK 和外部验证批次的所有测量的成功率均高于 0.8。Pearson 系数检验、SQ 和 ImageJ 之间的差异检验以及 SQ 和 ImageJ 之间的比例差异检验均在 0.05 水平上具有统计学意义。我们的结果表明，SQ 工具非常适合自动组织学染色图像定量分析。
