Manual microscopic analysis is the gold standard for analyzing blood smear. Microscopic analysis of blood smear by a hematologist is subjected to many challenges such as inter-observer variations, operator experience, and conditions of observation. This study aims to examine several parameters extracting from the features of blood smear images. These parameters were used to develop a predictive function, which can be used to automate the microscopic analysis of blood cells instead of manual observation. Fractal dimension, roundness, and aspect ratio were estimated for two types of abnormal erythrocytes: echinocyte and sickle cell. Standard conditions and the choosing of the optimum parameters through the imaging preprocessing were done in order to ensure that the chosen parameters reflect the morphological characteristics of examined erythrocytes. Statistical discriminant analysis was used to build the predictive function for erythrocytes morphological change by a linear combination of the measured parameters. The measured fractal dimensions were 1.825 ±0.008, 1.502 ±0.019 and 1.620 ±0.018 for control, echinocyte, and sickle cell, respectively. The roundness values were 0.94 ±0.05, 0.83 ±0.04 and 0.56 ±0.02 for control, echinocyte, and sickle cell, respectively. The aspect ratio values were 1.005 ±0.151, 1.046 ±0.089 and 1.742 ±0.162 for control, echinocyte, and sickle cell, respectively. The differences between the image analysis parameters for echinocyte and sickle, when compared to control, were statistically significant. The constructed discriminant function using measured parameters was effectively differentiating between examined erythrocytes. The results demonstrated that the selected image analysis parameters extracted from microscopic images with conjunction with statistical discriminant analysis could be used as powerful tools in the classification of erythrocytes according to their morphological characteristics. The findings of this study, in addition to the previous attempts in this filed, could help in the enhancement of a fully automated microscopic system for blood smear analysis.

中文翻译：

---

优化分形维数和形状分析作为判别红细胞异常的指标。可重现的诊断工具的新方法

手动显微镜分析是分析血液涂片的金标准。血液科医生对血液涂片的显微镜分析面临许多挑战，例如观察者之间的差异，操作者的经验和观察条件。这项研究旨在检查从血液涂片图像特征中提取的几个参数。这些参数用于开发预测功能，可用于自动进行血细胞的显微分析，而无需人工观察。估计了两种类型的异常红细胞的分形维数，圆度和纵横比：棘突细胞和镰状细胞。通过成像预处理进行标准条件和最佳参数的选择，以确保所选择的参数反映所检查的红细胞的形态特征。统计判别分析通过测量参数的线性组合用于建立红细胞形态变化的预测功能。对照，棘突细胞和镰状细胞的分形维数分别为1.825±0.008、1.502±0.019和1.620±0.018。对于对照，棘突细胞和镰状细胞，圆度值分别为0.94±0.05、0.83±0.04和0.56±0.02。对照，棘突细胞和镰状细胞的长宽比值分别为1.005±0.151、1.046±0.089和1.742±0.162。与对照相比，棘突肌和镰刀的图像分析参数之间的差异具有统计学意义。使用测量的参数构造的判别功能可以有效地区分所检查的红细胞。结果表明，从显微图像中提取的选定图像分析参数以及统计判别分析可作为根据形态特征进行红细胞分类的有力工具。除了本研究的先前尝试之外，本研究的发现还可以帮助增强用于血液涂片分析的全自动显微镜系统。