2022 Fiscal Year Final Research Report

Construction of Breast Cancer Tissue Image Analysis Algorithm Using Deep Machine Learning System with Artificial Intelligence

Research Project

PDF

Project/Area Number	20K07637
Research Category	Grant-in-Aid for Scientific Research (C)
Allocation Type	Multi-year Fund
Section	一般
Review Section	Basic Section 50020:Tumor diagnostics and therapeutics-related
Research Institution	Kyushu University
Principal Investigator	KAI Masaya 九州大学, 医学研究院, 共同研究員 (10755242)
Co-Investigator(Kenkyū-buntansha)	中村雅史九州大学, 医学研究院, 教授 (30372741) 久保真九州大学, 医学研究院, 准教授 (60403961) 小田義直九州大学, 医学研究院, 教授 (70291515) 中津川宗秀東京医科大学, 医学部, 教授 (70448596)
Project Period (FY)	2020-04-01 – 2023-03-31
Keywords	乳癌 / 人工知能 / 免疫染色 / 個別化医療
Outline of Final Research Achievements	In this study, by performing deep learning with a Convolutional Neural Network (CNN) using Whole Slide Image (WSI), ER,PgR,HER2,Ki67 was evaluated. We were evaluated 69.46% accuracy was obtained for ER, even with this number of cases. This indicates that the HE stained images can predict ER expression with high sensitivity and specificity. On the other hand, the accuracy of PgR and Ki67 was lower than that of ER. It was difficult to evaluate HER2 because it is a cell membrane staining and the algorithm is different from nuclear staining.In conclusion, the present study shows that some of the markers targeted as subtype determinants can be predicted from HE staining. For HER2, it was difficult to evaluate because it is a plasma membrane stain and the algorithm is different from that of nuclear staining such as ER, PgR, and Ki67. In conclusion, the present study shows that some of the markers targeted as subtype-specific factors can be predicted from HE staining.
Free Research Field	医歯薬学
Academic Significance and Societal Importance of the Research Achievements	乳癌においては、NSGによる生物学的特性の解析が発展し、今後はその特性に基づいた個別化医療が進められることが予想される。すなわち、膨大な臨床病理学的データをもとに、治療方針を判断しなければならなくなり、従来の人の手や目を中心とした解析作業には限界があると考えられるため、データを、正確・高速に処理し、診断から治療につなげるシステムの構築が急務であり、診断の効率化と個別化治療開発は、同時進行として進められるべき命題であると考えた。本研究では、AIによるDeep learningを行うことで、形態からタンパク発現の予測が、一定の精度と効率化をもって、診断につなげることができるということが示唆された。