Bakhshayesh Nayyer Mostaghim, Shamsi Mousa, Golabi Faegheh
Faculty of Biomedical Engineering, Sahand University of Technology, Tabriz, Iran.
Department of Advanced Medical Sciences, Tabriz University of Medical Sciences, Tabriz, Iran.
J Med Signals Sens. 2024 Mar 26;14:6. doi: 10.4103/jmss.jmss_65_22. eCollection 2024.
Microarray is a sophisticated tool that concurrently analyzes the expression levels of thousands of genes, giving scientists an overview of DNA and RNA study. This procedure is divided into three stages: contact with biological samples, data extraction, and data analysis. Because expression levels are disclosed by the interplay of light with fluorescent markers, the data extraction stage relies on image processing methods. To extract quantitative information from the microarray image (MAI), four steps of preprocessing, gridding, segmentation, and intensity quantification are required. During the generation of MAIs, a large number of error-prone processes occur, leading to structural problems and reduced quality in the resulting data, affecting the identification of expressed genes.
In this article, the first stage has been examined. In the preprocessing stage, the contrast of the images is first enhanced using the genetic algorithm, then the source noises that appear as small artifacts are removed using morphology, and finally, to confirm the effect of the contrast enhancement (CE) on the main stages of microarray data processing, gridding is checked on complementary deoxyribonucleic acid MAIs.
The comparison of the obtained results with an adaptive histogram equalization (AHE) and multi-decomposition histogram equalization (M-DHE) methods shows the superiority and efficiency of the proposed method. For example, the image contrast of the Genomic Medicine Research Center Laboratory dataset is 3.24, which is 42.91 with the proposed method and 13.48 and 32.40 with the AHE and M-DHE methods, respectively.
The performance of the proposed methods for CE is evaluated on 3 databases and a general conclusion is obtained as to which CE method is more suitable for each dataset.
微阵列是一种精密工具，可同时分析数千个基因的表达水平，使科学家对DNA和RNA研究有一个全面了解。该过程分为三个阶段：与生物样本接触、数据提取和数据分析。由于表达水平是通过光与荧光标记的相互作用来揭示的，因此数据提取阶段依赖于图像处理方法。要从微阵列图像（MAI）中提取定量信息，需要进行预处理、网格化、分割和强度量化四个步骤。在MAI生成过程中，会出现大量容易出错的过程，导致所得数据出现结构问题和质量下降，影响表达基因的识别。
在本文中，对第一阶段进行了研究。在预处理阶段，首先使用遗传算法增强图像对比度，然后使用形态学去除表现为小伪影的源噪声，最后，为了确认对比度增强（CE）对微阵列数据处理主要阶段的影响，在互补脱氧核糖核酸MAI上检查网格化。
将所得结果与自适应直方图均衡化（AHE）和多分解直方图均衡化（M-DHE）方法进行比较，显示了所提方法的优越性和有效性。例如，基因组医学研究中心实验室数据集的图像对比度为3.24，所提方法为42.91，AHE和M-DHE方法分别为13.48和32.40。
在所提的CE方法在3个数据库上进行了性能评估，并得出了关于哪种CE方法更适合每个数据集的一般性结论。
