The results of Olympus’ ongoing joint research programto create an AI-based pathology diagnostic tool with the potential to streamline pathologists’ workloads were announced at the recent Japan Society of Digital Pathology Study annual meeting. The diagnostic tool achieved 100% sensitivity and 50% or more specificity for all gastric biopsy pathology specimens analyzed.

Diagnostic Tool Demand

The ongoing shortage of pathologists has led to a demand for AI-based pathology diagnostic tools. Olympus, through its Office of Innovation*, began a collaboration with the Kure Medical Center and Chugoku Cancer Center in Japan in 2017 to develop an AI-based pathology diagnostic tool. In the initial testing phase, the AI was trained using 368 gastric biopsy pathology slide images.

The second phase of research5  began in November 2020, where the diagnostic tool was expanded to six hospitals in Japan, with the aim of verifying the versatility and improving the accuracy of the AI tool. Specifically, it was important to test whether the tool works correctly on pathology slides that vary in thickness and color.

The goal of this program is to deliver AI pathology diagnosis software that can assist pathologists by 2023.

Preliminary Results

The AI-based pathology tool uses a convolutional neural network (CNN) optimized to analyze pathology images. This technology enables the tool to identify adenocarcinoma versus non-adenocarcinoma tissue in an image. Once the AI was trained, it was tested using 1200 pathology whole slide images from the six institutions participating in the study. The AI classified each image as either adenocarcinoma or non-adenocarcinoma. The AI tool was able to achieve 100% sensitivity and 50% or higher specificity for slides from all six facilities. The robustness of the results will enable Olympus to pursue commercialization of the AI tool in the future.

For more information about Olympus life science products, please visit Olympus-LifeScience.com.

Featured Image: Areas with a high probability of adenocarcinoma tissue are displayed in red, and areas with a low probability of adenocarcinoma tissue are displayed in blue. Photo: Olympus

References

1National Hospital Organization Kure Medical Center and Chugoku Cancer Center, National Hospital Organization Osaka National Hospital, National Hospital Organization Shikoku Cancer Center, National Hospital Organization Nagasaki Medical Center, Hiroshima Memorial Hospital, and Kure Medical Association Hospital.

2Percentage of positive samples diagnosed as positive.

3Percentage of negative samples diagnosed as negative.

4Learn more about the first phase https://www.olympus-global.com/news/2018/nr00869.html

5Learn more about the second phase https://www.olympus-global.com/news/2020/nr01948.html

6A method widely used as a deep-learning technology for image analysis. The features to be analyzed can be learned efficiently.

7One of the cancers. Most cancers in gastric biopsy are said to be adenocarcinoma.

*Collaboration with companies and academic organizations with advanced technologies under our X (cross) INNOVATION program.