NCKU develops automatic pneumonia detection system for early detection of COVID-19 patients
Under the lack of doctors and nurses in countries with severe condition of COVID-19, pressure of taking care of patients and diagnosing symptoms are growing bigger. Before the outbreak of Coronavirus, researches on conducting artificial intelligent on healthcare system has been a key development to National Cheng Kung University (NCKU) Hospital, Taiwan.
In the past two years, department of diagnostic radiology of NCKU Hospital has been continuously conducting research on using AI in X-ray for faster chest diagnosis on pneumothorax and aortic dissection.
Using a dataset of chest X-rays, this artificial intelligence tool can alert doctors to a coronavirus infection in as little as a second.
With the existing research foundation on AI chest diagnostics, a team named MedCheX composed of NCKU researchers, Prof. Jung-Hsien Chiang, Dr. Yi-Shan Tsai, graduate and PhD students Chi-Siang Wang, Huang-Bin Chiou, Jhao-Yi Wu, is selected as one of the 89 highlighted projects out of 1560 teams around the world in the “Build for COVID19 Global Online Hackathon” competition. Tiffany Chiang, an elementary school teacher helped in language revision for the video used for competition.
MedChex is an e-Alert system for automatically detecting high-risk patients with pneumonia from chest x-rays. The system learn from the vast negative and positive x-ray image of COVID-19 provided by Dr. Yi-Shan Tsai, then detects pneumonia on x-rays images and identifies whether the person is infected by COVID-19.
The team combined ResNeXt with Feature Pyramid Network (FPN) based deep learning model to automatically detect the presence of pneumonia using this dataset. They also consulted with engineers in order to better integrate the model to assist doctors with their diagnoses. Transfer of x-ray scans, server construction, and prediction response were all needed to build this e-Alert system.
If a positive result is detected, doctors are notified of the suspected case via e-Alert (on computer or smartphone), and labeling of the corresponding area in the diseased lung are accomplished with a deep learning approach. The system’s accuracy can be up to 92%. With all the automatic process, doctors could have more accurate diagnosis and prompt treatments, which saves lots of human labor.
The system has already been tested with and adopted by doctors at the National Cheng Kung University Hospital. With this system, diagnosis could be finished within a second, minimizing time for patients waiting for the result and lowering the risk of infection in the hospital, making NCKU the first hospital in Taiwan to use AI technology in pneumonia diagnosis. Places insufficient with doctors would be hugely benefited from it.
In the future, other medical scans, such as CT scans or MRIs, may be included in this system to provide doctors with more accurate information to diagnose patients.
