The National Institute of Informatics, in collaboration with Nagoya University, Juntendo University, and the Japanese Society of Medical Radiology, has developed a platform for AI analysis of CT images of the new coronavirus pneumonia (COVID-19 pneumonia). It is expected that the accuracy of COVID-19 inspection will be improved.
RT-PCR is often used for COVID-19 testing, but its sensitivity (true positive rate) is 42% to 71%.On the other hand, the sensitivity of the COVID-19 test using CT images is as high as 97%.Since Japan has a large number of CT imaging devices per population, it can be expected to utilize CT images. The challenge is to establish an evaluation method that objectively judges the lung condition using a computer in order to support diagnosis using CT images and quantify the judgment.
In the development of this AI analysis platform, we have developed a machine learning method that selects pneumonia CT images from more than 1 million CT images collected and accumulated on the medical big data cloud platform of the National Institute of Informatics.The CT images of pneumonia listed by the method were added with the actual PCR test results and the judgment results of COVID-6 pneumonia typicality by a radiologist to create a database.In this way, we prepared a learning data set for AI and developed it as a platform for AI analysis for COVID-19 pneumonia research.
A research team at Nagoya University has independently applied AI to the CT image database of COVID-19 pneumonia cases to develop a highly accurate method for determining the typicality of COVID-19 pneumonia.This achieves 83.3% typicality discrimination performance.We also developed a method for AI to accurately estimate the shape of the lung when it is difficult to identify the lung on CT images due to the effects of inflammation and the like.
The research team aims to further improve the AI analysis platform for COVID-19 pneumonia research and improve the accuracy of AI selection and judgment by improving the AI algorithm in collaboration with other research teams.Furthermore, it is said to be effective in dealing with national urgent issues such as unknown infectious diseases other than COVID-19 in the future.
