Image data from multiple hospitals improves the tool’s versatility and accuracy.
Evident has recently announced the latest results from its ongoing joint research program to create an AI-based pathology diagnostic tool. The results show the diagnostic tool in development succeeded in expanding its AI detection to seven types of gastric cancer, highlighting its versatility for a range of pathology applications.
Evident, which is a wholly owned subsidiary of Olympus Corporation, unveiled the results at the 20th annual meeting of the Japanese Society of Digital Pathology.
There is an increasing demand for diagnostic tools that can help reduce the workload of pathologists. To meet this need, Evident began a collaboration with the National Hospital Organization Kure Medical Center and Chugoku Cancer Center in 2017 to develop an AI-based pathology diagnosis tool.
The first testing phase showed the diagnostic tool could successfully identify adenocarcinoma tissue from pathology whole-slide images. As part of the second research phase that began in November 2020, the tool was tested on 2,717 pathology whole-slide images provided by six hospitals in Japan to improve its versatility and accuracy.
By significantly increasing the amount of image data through collaboration with the hospitals, the diagnostic tool improved its AI detection of adenocarcinoma of tubular and poorly differentiated types and expanded its AI detection to identify other gastric cancers. It achieved a false negative rate of 0 to 2.5% in seven types of gastric cancer: adenocarcinoma, of tubular, papillary, mucinous and poorly differentiated types, gastrointestinal stromal tumour, MALT lymphoma, and diffuse large B-cell lymphoma. A common AI discrimination threshold was set for all hospitals, demonstrating the improved versatility of the software.
As the second research phase continues, Evident aims to further refine the tool to prepare it for commercial use. The goal of this program is to deliver an AI pathology diagnosis software that can assist pathologists by 2023.
For more information, visit EvidentScientific.com.