Research on the analysis of histopathologic images (Development of hisopathologic diagnosis support system for hepatocellular carcinoma)

本研究的目的是开发一个肝细胞癌组织病理学诊断支持系统。我们开发了一个专门用于非癌肝细胞癌和极高分化肝细胞癌鉴别诊断的支持系统。首先，提出了一种三步法来提取肝细胞核的位置。(一)候选人提取步骤：提取核职位的候选人。(Ii)轮廓提取步骤：开发并使用了一种新的轮廓提取方法，我们称之为“径向蛇法”。(Iii)筛选步骤：根据能量值去除大部分错误提取的位置。使用该方法，正确提取了84.8%的核位置，错误提取了12.2%的核位置。结果，使用图形用户界面纠正错误位置所需的时间大大减少。利用提取的位置计算核密度和高核密度面积比作为两个特征。这两种特征的敏感度和特异度均超过90%，表明这些特征对正常和新的肝癌的鉴别诊断是有效的。开发了一个图形用户界面，使用户能够通过简单的按钮操作来提取核位置、校正位置和计算所建议的特征。处理一幅图像和提取特征的总时间约为2分钟。该系统对实际诊断有一定的帮助。我们将改进这一系统，以便它能在更多的医疗机构中使用。

The purpose of this research was to develop a histopathologic diagnosis support system forhepatocellular carcinoma. We developed a support system especially for differential diagnosis between non-cancer and extremely well-differentiated hepatocellular carcinoma (ewHCC)First, a following three-step method to extract the positions of hepatocellular nuclei were developed. (i) Candidates extraction step: the candidates of nuclear positions are extracted. (ii)Contours extraction step: a novel contour extraction method, which we call "Radial Snakes" has been developed and used. (iii)Screening step: most of wrongly extracted positions are removed based on the energy values. Using this method, 84.8% of the nuclear positions were correctly extracted with 12.2% of the positions wrongly extracted. As a result, the time required to correct the wrong positions using a graphical user interface (GUI) was drastically reduced.Nuclear density and ratio of high nuclear density area are proposed as the two features calculated using the extracted positions. The average of the sensitivity and specificity was over 90% for both features, showing the effectiveness of these features for differential diagnosis between normal and ewHCC.A GUI was developed that enables a user to extract the nuclear positions, to correct the positions, and to calculate the proposed features by simple button operations. The total time for processing one image and extracting the features was about 2 minutes. This system would be helpful in actual diagnosis. We are going to improve the system so that it can be used in more medical organizations.