Nankai University and Infervision Use CT Image AI Screening to Help Prevent and Control of COVID-19

2020-03-17

AI technology is providing great support in the battle against the COVID-19 epidemic. Prof. Mingming Cheng from Nankai University’s College of Computer Science and Engineering and his team is collaboratively working with Infervision on COVID-19 CT image AI screening. 40 hospitals in China, including the ones in Hubei, started to use the system to assist doctors in the rapid diagnosis of novel coronavirus pneumonia, in the assessment of degree and in the dynamic monitoring of disease course. Till March 12, the system has been running for more than 50 days. It has cumulatively screened 81,000 cases and assisted doctors in the diagnosis of more than 6,000 cases of novel coronavirus pneumonia. The system’s degree of sensitivity (correct diagnosis rate) was 98.3%, and the degree of specificity (correct exclusion rate) was 81.7. %. The system is able to conclude the calculation of a case of 300 CT images in just about 10 seconds.

Doctors in Wuhan Tongji Hospital are using CT image AI screening system

In early 2019, Nankai University and Infervision has started related researches on the application of AI technology in CT image diagnosis.They have achieved a series of outstanding results in the identification of tuberculosis and detection of pulmonary nodules based on CT images.The large experimental data set that they built is more than 10 times the previous largest data set in the world. On this basis, they carried out relevant experiments, taking the confirmed data obtained six weeks after bacterial culture in the biosafety Level 3 laboratory as the “golden standard”, so that the average value of sensitivity and specificity of CT image screening based on AI technology exceeded the average accuracy of the judgment results of the chief doctors in the third grade-A hospital.

Infervision’s screenshot of AI diagnosis system based on lung CT image data

The CT imaging AI screening system for detecting COVID-19 includes two major modules. Among them, the imaging diagnosis module is mainly based on learning data of thousands of newly diagnosed cases of novel coronavirus pneumonia, assisting doctors to prompt suspected inflammation areas. The system has two main functions: one is rapid screening and early warning of novel coronavirus pneumonia, the other is digital accurate diagnosis and course monitoring.

The application of class activation map locates while classifying, 

providing better assistance for doctors

System operation interface of novel coronavirus pneumonia CT image AI screening

 “After the system is put into use, we will further accumulate clinical data during the operation and use this data to continuously improve the diagnostic performance of the system. We will improve the detection performance in the future so as to better assist doctors in making fast and accurate diagnosis. ” Prof. Mingming Cheng said.

(Reported by Junhui Wu, Translated by Yuchen Shi, Edited by Davide Francolino and JianjingYun)