成果報告書詳細
管理番号20120000001176
タイトル*平成23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)」(2)
公開日2012/12/19
報告書年度2011-2011
委託先名国立大学法人東京工業大学
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
和文要約和文要約等以下本編抜粋:
1. 研究開発の内容及び成果等
(1) 画像特徴量を用いた数量化技術の開発
(1-1) 類洞抽出法の開発
肝細胞がんにおける病理組織の構造的特徴を数量化するための画像解析技術の開発を行った。がんによる構造異型の特徴は肝細胞の索状配列に現れるが、HE染色の病理標本の画像から肝細胞索構造を直接抽出する技術は確立していない。本研究では、肝細胞索の形態学的な特徴抽出の一環として索構造に栄養を送るために並走する類洞構造を抽出した。
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-FY2012) FY2011 Annual Report (Tokyo Institute of Technology)
This work focuses on two subjects; the quantification of image features and the accuracy nhancement of color and spectral information. In the research for image feature quantification, the HE-stained tissue images of hepatocellular carcinoma (HCC) were experimentally investigated. For understanding the structure of hepatocytes, a method for the segmentation of hepatic sinusoids in HE-stained liver specimens is proposed. In the method, a new orientation selective filter for edge enhancement and de-noising is applied as a preprocessing. Then a clustering technique is applied to image pixels for identifying candidate sinusoids, which is followed by a classification procedure for discarding unlikely candidate and selecting final sinusoidal region. Experimental results showed the effectiveness of the proposed method for extracting sinusoidal region and the possibility of characterizing the grade or differentiation of hepatocytes using the structure of the extracted sinusoids. As another approach of quantifying the feature of the liver tissue morphology, we exploited the multifractal analysis to represent the histologic texture. In the experiment, ROI images of 2174x2174 pixels stored in the database were used. Then 50 small patches (256x256 pixels) were randomly selected from the ROI image and classified into the different grades by SVM (support vector machine). The Edmondson Grade of each ROI was determined by majority voting. The overall classification rate was very high, i.e., about 95%. For the accuracy enhancement of color and spectral information, we have developed a color correction method for whole slide image scanner using the color chart slide prepared by chusetts General Hospital. The developed method was applied to digital slide scanners, and experimentally evaluated by comparing with the true color image btained from the spectral imaging system. As a result, the color difference (delta-E*ab) between the corrected image and the true color image was improved to 8.80, while it was 13.46 before the color correction. In addition to the correction of device dependencies, the color variation due to the different staining condition should be also compensated. In this work, a color conversion method for compensating staining variation was developed and implemented in the system of liver fibrosis quantification with using Elastica van Gieson (EVG) staining. It enables the system to process the slides with color variation tomatically. Multispectral imaging technique is valuable for the evaluation of color reproducibility as well as the exploitation of spectral features. The system for spectral imaging with image tiling has been improved by incorporating automatic focus adjustment. Finally, as an application of spectral imaging, digital staining technique is successfully applied to immunohistochemistry stained slides.
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