Bartels P H, Thompson D, Montironi R
Optical Sciences Center, University of Arizona, Tucson 85721, USA.
Eur Urol. 1996;30(2):234-42. doi: 10.1159/000474174.
It is the objective of this study to present the use of knowledge-guided procedures in quantitative image analysis and interpretation in histopathology.
The knowledge-guided procedures were implemented in the form of N-gram encoding methods for the search and detection of areas of atypicality or abnormality in histopathologic sections; they were implemented as expert system for automated scene segmentation based on an associative network with frames at each node. The extraction of histometric features from the basal cell layer of prostatic lesions is presented as an example of automated image interpretation.
Rapid search algorithms for lesion detection were able to identify approximately 90% of areas labelled as atypical or abnormal by visual assessment, in lesions of colon, prostate and breast. Automated segmentation of very complex histopathologic imagery was possible with a success rate of approximately 80-90%, in sections of prostatic and colonic lesions. Histometry of the deterioration of the basal cell layer in prostatic lesions provided a monotonic trend curve suitable for the measurement of progression or regression.
Knowledge-guided procedures bring external information, not offered by the imagery itself, to bear on image processing and image analytic methods. This has enabled automated analysis and interpretation of very complex imagery, such as from cribriform glands, resulting in quantitative diagnostic information.
本研究的目的是介绍知识引导程序在组织病理学定量图像分析和解读中的应用。
知识引导程序以N-gram编码方法的形式实现，用于在组织病理学切片中搜索和检测非典型或异常区域；它们被实现为基于每个节点带有框架的关联网络的自动场景分割专家系统。以从前列腺病变的基底细胞层提取组织计量学特征为例，展示自动图像解读。
在结肠、前列腺和乳腺病变中，用于病变检测的快速搜索算法能够识别出约90%经视觉评估标记为非典型或异常的区域。在前列腺和结肠病变切片中，能够对非常复杂的组织病理学图像进行自动分割，成功率约为80%-90%。前列腺病变基底细胞层退变的组织计量学提供了一条适合测量进展或消退的单调趋势曲线。
知识引导程序将图像本身未提供的外部信息应用于图像处理和图像分析方法。这使得能够对非常复杂的图像（如筛状腺体图像）进行自动分析和解读，从而获得定量诊断信息。
