Development of deep learning algorithms to detect minor lymph node metastases of gastric cancer for histopathologic specimens

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K09027
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 55020:Digestive surgery-related
• Research Institution: Chiba University
• Principal Investigator: Hayashi Hideki 千葉大学, フロンティア医工学センター, 教授 (20312960)
• Co-Investigator(Kenkyū-buntansha): 吉村 裕一郎 富山大学, 学術研究部医学系, 特命助教 (90826471)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 人工知能 / 病理診断補助 / ResNet-152 / 転移学習 / 臨床試験

Outline of Final Research Achievements

We developed algorithms to detect histological metastasis of gastric cacer on a disitized regional lymph nodes images obtained from cases with gastric cancer who underwent surgery for curative intent. We tested performances of various machine learning techniques to detect metastases, and ResNet-152 pretrained with ImageNet showed the best. Therefore, we conducted the clinical trial to evaluated its utility in the practical pathological workflow. Consequently, no statistical differences were observed in accuracy and diagnostic time for metastasis negative, isolated tumor cell, and macrometastasis; however, statistically significant improvement of diagnostic accuracy was observed for micrometastasis, although its diagnositic time was significantly exteded.

Free Research Field

消化器外科学

Academic Significance and Societal Importance of the Research Achievements

胃癌のリンパ節転移診断において、現在最も正確性の高い機械学習の手法を明らかにした。この手法を用いた補助的病理診断が、特に微小転移（転移巣の大きさが0.2 - 2mm）の診断に貢献することが明らかとなった。本研究を通じて、人工知能が病理組織診断においてどのような貢献が期待できるかを明らかにすると共に、機械学習用いた病理診断の限界と、人間の行う病理診断の限界の違いも明らかにされたことから、今後の消化器癌の病理組織診断における人工知能開発の方向性に指針を与えるものと期待される。