成果報告書詳細
管理番号20140000000458
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 病理画像等認識自動化システムの研究開発 (定量的病理診断を可能とする病理画像解析システム)」
公開日2014/7/26
報告書年度2013 - 2013
委託先名日本電気株式会社
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We have been developing a hepatocellular carcinoma (HCC) detection program on the basis of nuclear features extracted from images of surgery tissue sections. In this fiscal year, we evaluated the accuracy of the HCC detection program using liver biopsy images. In this evaluation, we divided the original ROI (size of 2048x2048 pixels) into 4 sub-ROIs (size of1024x1024 pixels) and employed three types of total decision rules to determine HCC positive/negative for the original ROI level: (1) majority vote rule, i.e., positive if majority of sub-ROIs were positive, otherwise negative, (2) aggressive rule, i.e., positive if one or more sub-ROIs were positive, otherwise negative, and (3)conservative rule, i.e., negative if one or more sub-ROIs were negative, otherwise positive. The false negative (FN) / false positive(FP) rates were 3.7/63.9%, 3.3/75.0%, and 9.8/36.9%, respectively. The best accuracy was obtained by the conservative rule and this accuracy was promising considering that the detection was done on the basis of nuclear features only. To boost the accuracy of HCC detection and provide pathologists with quantitative information of liver biopsy tissue, we developed a prototype of liver biopsy feature extraction software by compiling all algorithms developed in this project. During next fiscal year, we plan to conduct a clinical evaluation of this prototype using liver biopsy images collected at the Department of Pathology of Keio University.
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