Biopharmaceutical Bioprocessing Technology Centre, Chemical Engineering and Advanced Materials, Newcastle University, Newcastle-upon-Tyne, UK.
BMC Med Imaging. 2014 Feb 12;14:7. doi: 10.1186/1471-2342-14-7.
Digital image analysis has the potential to address issues surrounding traditional histological techniques including a lack of objectivity and high variability, through the application of quantitative analysis. A key initial step in image analysis is the identification of regions of interest. A widely applied methodology is that of segmentation. This paper proposes the application of image analysis techniques to segment skin tissue with varying degrees of histopathological damage. The segmentation of human tissue is challenging as a consequence of the complexity of the tissue structures and inconsistencies in tissue preparation, hence there is a need for a new robust method with the capability to handle the additional challenges materialising from histopathological damage.
A new algorithm has been developed which combines enhanced colour information, created following a transformation to the Lab* colourspace, with general image intensity information. A colour normalisation step is included to enhance the algorithm's robustness to variations in the lighting and staining of the input images. The resulting optimised image is subjected to thresholding and the segmentation is fine-tuned using a combination of morphological processing and object classification rules. The segmentation algorithm was tested on 40 digital images of haematoxylin & eosin (H&E) stained skin biopsies. Accuracy, sensitivity and specificity of the algorithmic procedure were assessed through the comparison of the proposed methodology against manual methods.
Experimental results show the proposed fully automated methodology segments the epidermis with a mean specificity of 97.7%, a mean sensitivity of 89.4% and a mean accuracy of 96.5%. When a simple user interaction step is included, the specificity increases to 98.0%, the sensitivity to 91.0% and the accuracy to 96.8%. The algorithm segments effectively for different severities of tissue damage.
Epidermal segmentation is a crucial first step in a range of applications including melanoma detection and the assessment of histopathological damage in skin. The proposed methodology is able to segment the epidermis with different levels of histological damage. The basic method framework could be applied to segmentation of other epithelial tissues.
数字图像分析具有解决传统组织学技术问题的潜力，包括通过应用定量分析来解决缺乏客观性和高变异性的问题。图像分析的一个关键初始步骤是确定感兴趣区域。一种广泛应用的方法是分割。本文提出应用图像分析技术对具有不同程度组织病理学损伤的皮肤组织进行分割。由于组织结构的复杂性和组织准备的不一致性，人体组织的分割具有挑战性，因此需要一种新的稳健方法，该方法具有处理组织病理学损伤带来的额外挑战的能力。
我们开发了一种新算法，该算法结合了增强的颜色信息，这些信息是在 Lab*颜色空间转换后创建的，同时还结合了一般的图像强度信息。包括颜色归一化步骤，以增强算法对输入图像的照明和染色变化的鲁棒性。对优化后的图像进行阈值处理，并使用形态学处理和对象分类规则的组合对分割进行微调。将分割算法应用于 40 张苏木精和伊红（H&E）染色皮肤活检的数字图像上。通过将提出的方法与手动方法进行比较，评估算法程序的准确性、灵敏度和特异性。
实验结果表明，所提出的全自动方法对表皮的分割具有 97.7%的平均特异性、89.4%的平均灵敏度和 96.5%的平均准确性。当包括一个简单的用户交互步骤时，特异性提高到 98.0%，灵敏度提高到 91.0%，准确性提高到 96.8%。该算法能够有效地对不同严重程度的组织损伤进行分割。
表皮分割是包括黑色素瘤检测和皮肤组织病理学损伤评估在内的一系列应用的关键第一步。所提出的方法能够对具有不同水平组织学损伤的表皮进行分割。基本方法框架可应用于其他上皮组织的分割。
