Summary of Research Results
Successfully developed artificial intelligence to detect invasive ductal carcinoma in digital mammary histopathology specimens.
Background of this study
This research is the result of a developmental application of our research on "Creation of Basic Technology for Transfer Learning," which was presented internationally at the MIDL Conference in July 2021 and published in a paper. In Japan, the incidence of breast cancer continues to rise, and according to the "Latest Cancer Statistics," the lifetime incidence rate for women has exceeded 10%. In addition to the diagnosis of benign or malignant disease by biopsy, the importance of pathological diagnosis in the breast field is extremely high, including rapid intraoperative pathological diagnosis and the search for evaluation items that are important in determining treatment strategy.
The purpose of this study is to develop an artificial intelligence that can detect invasive ductal carcinoma (IDC) in digital mammary histopathology specimens by using deep learning.
Contents of this study
In this study, we developed an artificial intelligence to detect invasive ductal carcinoma of the breast without using a large amount of annotated data from multiple pathologists by performing transfer learning using the partial fine-tuning method (Proceedings of Machine Learning Research 143:338-353, 2021) and weakly-supervised learning (weakly-supervised learning). -supervised learning), we developed an artificial intelligence that can detect invasive ductal carcinoma of the breast without using a large amount of annotated data from multiple pathologists. In addition, the developed artificial intelligence was validated using validation cases that were different from the supervised data to verify its accuracy.
Results of this study
The developed model was validated on biopsy specimens, surgical specimens, and specimens from a public database (TCGA), and extremely high accuracy results with ROC-AUC of 0.95~0.98 were obtained in all validation cases. In addition, the areas suggesting invasive ductal carcinoma detected by the artificial intelligence displayed by the heat map were validated by a pathologist.

The method of developing the existing models published by MedMain has been innovatively developed, and we have succeeded in developing an efficient and highly accurate deep learning artificial intelligence by only transfer learning from existing models and assigning classification information (diagnostic information) at the sample level, without using a large amount of image features obtained from annotation data by detailed pathologists. We have succeeded in developing an efficient and highly accurate deep learning artificial intelligence without using a large amount of image features obtained from annotated data by detailed pathologists.
We will further validate the artificial intelligence model developed this time at multiple facilities and in large-scale cases to advance validation and basic research and development in transfer learning.
Original Papers
▼ Paper title: Breast invasive ductal carcinoma classification on whole slide images with weakly-supervised and transfer learning
▼ Japanese Translation: Development of Artificial Intelligence Using Deep Learning to Detect "Invasive Ductal Carcinoma of the Breast" in Digital Histopathology Specimens
▼ DOI:https://doi.org/10.3390/cancers13215368
■Author / Affiliation
<Medmain Inc.
Masayuki Tsuneki, Fahdi Kanavati
About Medmain Corporation
| name of company | Medmain Inc. |
| Establishment | January 11, 2018 |
| capital stock | [Tokyo Office] A Aoyama Bldg. 2F, 2-10-11 Minami-Aoyama, Minato-ku, Tokyo [Fukuoka Office] Chatelet Saxies 104, 2-4-5 Akasaka, Chuo-ku, Fukuoka City, Fukuoka Prefecture, Japan |
| representative | Tadashi Iizuka |
| Business Overview | Planning, development, operation and sales of medical software and cloud services |
| uniform resouce locator | https://www.j-startup.go.jp/startups/ Various related sites Corporate websitehttps://medmain.com/ Product Sitehttps://medmain.com/products Digital pathology support solution "PidPort Imaging Center" - High-quality digitization service for pathology specimens |









