Successful Development of an Artificial Intelligence Model Using Deep Learning Approach to Establish Histological Sub-classification of Lung Cancer in Biopsy Specimens Whole Slide Images - Published in Scientific Reports -
Medmain Co., Ltd. (Headquarters: Fukuoka Prefecture, Fukuoka City, CEO: Osamu Iizuka, hereinafter referred to “Medmain”), a provider of “PidPort” pathological diagnosis support solutions, has succeeded in developing an artificial intelligence model to classify lung cancer subtypes in lung biopsy specimens using Deep Learning, which enables the sub-classification of the macarcinoma). A dissertation on this development was jor histological subtypes (adenocarcinoma, squamous cell carcinoma and small-cell published on April 14, 2021 in “Scientific Reports”, a natural science journal issued by Nature Research in the United Kingdom.
■Outline of the Research Results
We successfully developed an artificial intelligence model to classify adenocarcinoma, squamous cell carcinoma, small-cell carcinoma and non-neoplastic lesions in TBLB (transbronchial lung biopsy) histopathologic specimens.
■Background of the Results
When deciding on a treatment for lung cancer, clinicians perform TBLB (transbronchial lung biopsy) for one of the important diagnostic aids. TBLB is a method of biopsy using a bronchoscope and forceps in the bronchus. The results of TBLB reveal the histopathological subtypes of lung cancer and/or genetic mutation data to determine the appropriate treatment procedure(s).
The purpose of this study is to develop an artificial intelligence model that can classify the histopathological subtypes of lung cancer in digital TBLB whole slide images (WSIs) by using deep learning approaches.
■Methods
In this study, we received TBLB histopathological specimens and pathological diagnoses from the National Hospital Organization Kyushu Medical Center and International University of Health and Welfare Mita Hospital. After digitizing the TBLB specimens, we prepared training sets for deep learning, and developed an artificial intelligence model by deep learning approaches. The accuracy of the established a deep learning model was validated using test sets from Kyushu Medical Center, Mita Hospital, and the International Database (TCGA).
■Results
In both TBLB and surgical cases, the ROC-AUC exhibited 0.94 or higher, which is an extremely high accuracy. Furthermore, when the results were validated on 83 TBLB specimens which were diagnosed based on immunohistochemistry in addition to conventional HE staining slides, ROC-AUC was 0.99. Based on the above, we have succeeded in developing an artificial intelligence model that can accurately classify adenocarcinoma, squamous cell carcinoma, small cell carcinoma, and non-neoplastic lesions in a diverse range of histopathological specimens. As a result, we are now able to classify lung cancer subtypes in more detail following the paper published in June 2020.
The accuracy of the newly developed artificial intelligence model will be verified at multiple facilities in the future
■Original Article
The results of this research were published in the online edition of Scientific Reports (April 14, 2021).
▼Thesis Title: A deep learning model for the classification of indeterminate lung carcinoma in biopsy whole slide images.
▼Japanese Translation: 肺癌生検病理組織デジタル標本における組織型分類を可能にする深層学習を用いた人工知能の開発
▼DOI：https://www.nature.com/articles/s41598-021-87644-7
■Author・Belonging Department
< Respiratory Surgery at Kyushu Medical Center>
Gouji Toyokawa, Koji Yamazaki, Sadanori Takeo
Seiya Momosaki
< Respiratory Medicine at Kyushu Medical Center>
Hiroaki Takeoka, Masaki Okamoto
Fahdi Kanavati, Osamu Iizuka, Masayuki Tsuneki
■Company Overview
【Company Name】Medmain Inc.
*Ministry of Economy, Trade and Industry J-START UP, Selected Company https://www.j-startup.go.jp/startups/
【Date of Establishment】01/11/2018
【Profile】Medical Software, Planning・Development・Operation and Sales of Cloud Services
【CEO】Osamu Iizuka
【Address】【Tokyo Office】2-10-11 Minami Aoyama #A Aoyama Bldg. 2F, Minato-ku, Tokyo 【Fukuoka Office】2-4-5 Akasaka #104, Chuo-ku, Fukuoka
■Related Sites
【Official Corporate Website】 https://medmain.com/
【Pathology Diagnosis Support Solution「PidPort」】 https://pidport.medmain.com/
【Imaging Center | Digitization Services for Pathology Specimens】 https://imaging.medmain.com/
■Contact us
Medmain Inc.
Sato (Mr.) at PR Dept. ： pr-m@medmain.com
Tweet