Institute for Biomedical Technologies of the National Research Council, Segrate, Milan, Italy.
BMC Bioinformatics. 2010 Nov 18;11:566. doi: 10.1186/1471-2105-11-566.
Tissue MicroArray technology aims to perform immunohistochemical staining on hundreds of different tissue samples simultaneously. It allows faster analysis, considerably reducing costs incurred in staining. A time consuming phase of the methodology is the selection of tissue areas within paraffin blocks: no utilities have been developed for the identification of areas to be punched from the donor block and assembled in the recipient block.
The presented work supports, in the specific case of a primary subtype of breast cancer (tubular breast cancer), the semi-automatic discrimination and localization between normal and pathological regions within the tissues. The diagnosis is performed by analysing specific morphological features of the sample such as the absence of a double layer of cells around the lumen and the decay of a regular glands-and-lobules structure. These features are analysed using an algorithm which performs the extraction of morphological parameters from images and compares them to experimentally validated threshold values. Results are satisfactory since in most of the cases the automatic diagnosis matches the response of the pathologists. In particular, on a total of 1296 sub-images showing normal and pathological areas of breast specimens, algorithm accuracy, sensitivity and specificity are respectively 89%, 84% and 94%.
The proposed work is a first attempt to demonstrate that automation in the Tissue MicroArray field is feasible and it can represent an important tool for scientists to cope with this high-throughput technique.
组织微阵列技术旨在同时对数百种不同的组织样本进行免疫组织化学染色。它允许更快地分析，大大降低了染色的成本。该方法学中一个耗时的阶段是选择石蜡块内的组织区域：尚未开发出用于从供体块中识别要打孔并组装到受体块中的区域的实用程序。
所提出的工作支持在管状乳腺癌的原发性亚型的特定情况下，在组织内的正常和病理区域之间进行半自动区分和定位。通过分析样本的特定形态特征（例如缺乏围绕腔的双层细胞以及规则腺体和小叶结构的退化）来进行诊断。使用一种算法来分析这些特征，该算法从图像中提取形态参数，并将其与经过实验验证的阈值进行比较。结果令人满意，因为在大多数情况下，自动诊断与病理学家的反应相匹配。特别是在总共 1296 个显示乳腺标本正常和病理区域的子图像中，算法的准确性、灵敏度和特异性分别为 89%、84%和 94%。
这项工作是首次尝试证明组织微阵列领域的自动化是可行的，它可以成为科学家应对这种高通量技术的重要工具。
