Medicine -Realization of Social Implementation of Next-generation Medicine Through Advanced Image and Video Recognition Technology-

As the first country in the world to enter the super-aging society, Japan is facing the chronic issues of shortage of doctors and increase in medical costs. Therefore, there is an urgent need to improve the efficiency of medical care by establishing diagnosis support technology based on machine learning. Although there are high expectations for the application of AI technology in medicine, it is necessary to overcome various issues such as improving the reliability of AI, securing the amount of training data, and ensuring security and anonymity to introduce AI into clinical settings. In Laboratory of Media Dynamics, we are challenging to solve the remaining problems in the medical field by using advanced image/video recognition technology as the core technology. We are committed to being world-class in the social implementation of next-generation medicine based on AI.

High Accuracy Gastric Cancer Risk Identification AI

 Gastric cancer is one of the most deadly types of cancer. The main cause is atrophy of the stomach due to H. pylori infection, and their stomachs inhabited by H. pylori have an increased risk of gastric cancer. Chronic atrophic gastritis is a disease with a high risk of gastric cancer, and diagnosis by X-ray imaging requires a high level of skill and experience.

In Laboratory of Media Dynamics, we have developed an AI system that can automatically determine the risk of gastric cancer. It has been reported that the judgment accuracy by doctors is about 90%, but the AI constructed in this study achieves a judgment accuracy of more than 96%, which is higher than that of doctors.

Visualization of the Basis for AI Decisions

It is important to ensure the reliability of the judgment results for introducing AI in the medical field. For example, even if an AI can make judgments in high accuracy, it is not a reliable AI if the basis for the judgments is inconsistent with that of the doctor.
In Laboratory of Media Dynamics, we have constructed ExplainableAI (XAI), which visualizes the regions of the image where AI focused on, to judge the risk of gastric cancer. The red areas shown in the figure below are the areas that the AI considers important when making judgments. On the other hand, the blue areas are not affected by the judgments. The red area is located in the inner region of the stomach. The XAI is expected to contribute to improving the reliability of AI, since doctors can judge the reliability of the AI by referring to the visualization results.

Efficient Training of AI using Generated Images

In the medical field, it is difficult to make medical image data publicly available for privacy protection reasons. Therefore, AI may have to be constructed from only independent data at each medical facility.
In Laboratory of Media Dynamics, we have constructed an AI system that can automatically generate images with disease-related findings using a technology called Generative Adversarial Network (GAN). The left side of the figure below shows an actual gastric X-ray image that represents the characteristics of chronic atrophic gastritis, and the right side shows the image generated by the AI. We have shown that AI can be trained efficiently by using generated images.

Global Attention and Technology Deployment

Introduced in Journal of Scientific American ( link )

Medical image distillation technology developed by Laboratory of Media Dynamics, which enables overwhelming reduction of the amount of data for AI training, was introduced in Scientific American, the oldest journal of general science in the world.

Expanding AI to Other Medical Image

We are working on the horizontal deployment of fundamental technologies developed by Laboratory of Media Dynamics. In joint research with Olympus Corporation, we are conducting research on endoscopic images. In collaboration with Hokkaido University School of Medicine, we are conducting research on PET/CT images.

Participation in international medical conferences

In order to conduct advanced research, it is also necessary to keep abreast of medical knowledge and trends in the field. We actively submit and participate in international medical conferences.

Cooperative Institutions

  • Olympus Corporation.
  • Junpukai Health Maintenance Center.
  • Hokkaido University School of Medicine.
  • Tokyo University School of Medicine.
  • Hyogo College of Medicine.
  • Hakodate National Hospital.
  • Miyagi Cancer Society.
  • Yamagata Medical Association.
  • Social Insurance Shiga Hospital.
  • NTT East Sapporo Hospital.