News Releases & Research Results Discovery of unknown cancer features with AI: self-acquisition of novel information on cancer recurrence

Results of R&D carried out by a collaborative research led by Team Leader Yoichiro Yamamoto of the Pathology Informatics Team, RIKEN Center for Advanced Intelligence Project, and Associate Professor Go Kimura of Department of Urology, Nippon Medical School Hospital.

• With an AI-based technology developed for autonomous acquisition of information on cancer from diagnostic annotation-free histopathology images, new features were successfully identified to allow the improvement of the accuracy of cancer recurrence diagnosis.
• Specifically, an AI-based technology was successfully developed to automatically obtain cancer features from prostate pathology images consisting of 10 billion pixels without prior training and report the results in form understandable to humans. Features identified by AI included cancer diagnostic criteria that have been used worldwide, as well as features in non-cancer areas that experts were not aware of.
• These results are expected to contribute to personalized medicine by providing a highly accurate prediction of cancer recurrence after surgery, as well as an automated analysis method to obtain new information from images.

This R&D project was conducted with the support of Project for Research on Creation of Medical Arts by AMED.

The research results were published in Nature Communications, a British scientific journal available online, on December 18.

Yamamoto Y., et al. Automated acquisition of explainable knowledge from unannotated histopathology images　Nature Communications

DOI：10.1038/s41467-019-13647-8