Huawei Launches New AI to Diagnose Coronavirus within seconds

Huawei Launches New AI to Diagnose Coronavirus within seconds


Islamabad 19th March 2020: In the face of the COVID-19 epidemic, the National Health Commission of China has included CT features of COVID-19 as the clinical diagnosis standard for COVID-19 in Hubei Province in its official diagnosis and treatment solution released on February 4, 2020. As one of the decision-making methods for COVID-19 diagnosis and treatment, CT powers fast and effective diagnosis and evaluation. However, due to large numbers of lesions in lungs and rapid changes, multiple rechecks and image reviews are required in a short period of time, significantly increasing the workloads of imaging doctors. What's worse, imaging doctors who can accurately diagnose and quantitatively analyze COVID-19 are insufficient in number, so the diagnosis efficiency cannot be greatly improved. 

Recently, HUAWEI CLOUD worked with Huazhong University of Science & Technology and Lanwon Technology to develop and launch an AI-assisted quantitative medical image analysis service for COVID-19. With HUAWEI CLOUD's leading AI technologies such as computer vision and medical image analysis, the service can automatically, quickly, and correctly output CT quantification results to imaging and clinical doctors, alleviating the shortage of imaging doctors who can accurately diagnose COVID-19, relieving the pressures of quarantines, and reducing the heavy workloads of doctors. The service also utilizes the powerful computing capability of Huawei Ascend series AI chips to output the quantization result of a single case in seconds. "AI + doctor review" is dozens of times faster than manual quantitative image evaluation, which greatly improves the diagnosis efficiency. 

HUAWEI CLOUD leverages the computer vision and medical image analysis technologies to segment the multiple pulmonary ground glass opacities (CGOs) and lung consolidation and make quantitative evaluations through CTs of patients' lungs. It combines clinical information and laboratory results to help doctors more accurately distinguish between early, advanced, and severe stages of COVID-19, facilitating early screening and prevention/control. For confirmed cases in hospitals, this AI-assist service can perform registration and quantitative analysis on the 4D dynamic data of multiple rechecks within a short period of time, helping doctors effectively evaluate patients' conditions and drug use effects. 

According to the analysis of hundreds of COVID-19 cases and non-infected cases, HUAWEI CLOUD AI-assisted quantitative diagnosis service has implemented industry-leading Dice Similarity Coefficient (DICE: overlaps between the predicted lesions and the actual lesions) and Absolute Volume Difference (AVD: volume differences between the predicted lesions and the actual lesions) of lesion area segmentation. The results are highly consistent with the precise manual sketching by doctors. In addition, with Huawei's Ascend AI series chips, the service can output CT quantization results in seconds. Compared with the traditional methods that doctors use to manually draw ROIs for quantitative evaluation, this service significantly improves diagnosis efficiency. 

The "AI + CT" medical image analysis service for COVID-19 is developed and constructed based on the HUAWEI CLOUD EIHealth-medical image analysis platform. The platform uses the HUAWEI CLOUD AI Ascend cluster service and the ModelArts one-stop AI development and management platform to implement one-stop medical image data governance and labeling, model training and evaluation, and visualized rendering, providing powerful support for universities and hospitals with massive AI computing, platform, and algorithm power. 

Recently, HUAWEI CLOUD has been working with various industries to carry out multiple AI-based scientific research projects on COVID-19. HUAWEI CLOUD has released the results of ultra-large-scale computer-assisted drug screening together with many scientific research institutions



Post a Comment

أحدث أقدم