成果報告書詳細
管理番号20140000000462
タイトル*平成25年度中間年報 「がん超早期診断・治療機器の総合研究開発 超早期高精度診断システムの研究開発:病理画像等認識技術の研究開発 病理画像等認識基礎技術の研究開発(定量的病理診断を可能とする病理画像認識技術)」
公開日2014/7/26
報告書年度2013 - 2013
委託先名国立大学法人東京工業大学 学校法人慶應義塾 学校法人埼玉医科大学 日本電気株式会社
プロジェクト番号P10003
部署名バイオテクノロジー・医療技術部
和文要約
英文要約Title: Research and Development Project for pathological image recognition technology, Research and Development Project to develop basic technologies for recognizing pathology images, Pathological image analysis technology to enable a quantitative pathological diagnosis, (FY2010-FY2014) FY2013 Annual Report
(Tokyo Institute of Technology, Keio University, Saitama Medical School, NEC Corporation)

1. Collecting the pathology information necessary to early cancer diagnosis:
In order to create the prototype software, we collected the whole slide images of the liver biopsy specimens of hepatocellular carcinoma (HCC). We selected candidate for morphological information of the cancer cells to predict the recurrence, portal invasion and intrahepatic metastasis of HCC and created the discriminants.
2. Development of a diagnostic marker quantification method:
We developed the molecular quantification technology on tissue sections. In the clinical trial, the expression of Glypican 3 in HCC was assessed by immunohistochemical staining used for DAB and immunofluorescent quantification digital slide (IQD). The relation of the molecular marker expression based on these data and the therapeutic effect of the molecular target drug against Glypican 3 was investigated. For DAB staining specimens of novel molecular markers, which were difficult to quantify, we attempted to quantify it by image processing. Furthermore, IQD was applied to the investigation of relationship between morphology and molecular expression.
3. Development of image recognition and quantification techniques:
We developed new features related to tissue structure; the irregularity of trabecular and the evenness of nuclei distribution. It was confirmed that the accuracy of HCC detection was improved by introducing these features addition to the features of nuclei. We also developed the features representing the change of tissue state: fatty cells and clear cells. The programs to calculate all the features have been implemented as a form that can be installed to the prototype system.
In addition, for high quality of quantitative diagnosis, we have developed extraction method of fat drops using the feature values of color information and arrangement of cell nuclei as well as shape information. We achieved accurate detection of fat droplets and quantify the fat droplet ratio against the image. As a quantitative evaluation characteristics using graph structure, we had proposed a high speed measuring algorithm to calculate the numbers of cell layers of hepatic trabecular.
4. Accuracy enhancement in pathology imaging:
A color correction method for HE-stained liver biopsy images was developed for stabilizing image analysis. This method performs the color correction so as to fit the color distribution of a target image to that of the reference image in a logarithm RGB color space. It was confirmed that the performance of the extraction of nuclei was improved by applying the proposed color correction before the extraction.
5. Algorithm evaluation:
In order to show the practical application of the system for histologically assessing liver fibrosis, evaluation of clinical validity was carried out. In our cases, the relationship between the liver stiffness measured by transient elastography and the median values of collagen fiber measured by our system was significant. We developed the prototype software equipped with this algorithm and investigated the correlation between fibrosis index and cancer risk.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る