U of G Researchers Use AI for Rapid Bedside COVID-19 Diagnosis

Facebooktwitterpinterestlinkedintumblrmail

Faster COVID-19 test results are the goal of a new automated diagnostic tool developed by University of Guelph researchers.

Dr. Eranga Ukwatta

Current diagnostic tests based on PCR (polymerase chain reaction) require off-site laboratory analysis. This takes time, and many tests are being ordered every day. Medical professionals can provide a bedside diagnosis of the disease with chest X-rays and ultrasound imaging, but this requires trained personnel and is subject to human error.

To shorten test result wait times, Dr. Eranga Ukwatta, professor in the School of Engineering, and Dr. Andrew Hamilton-Wright, professor in the School of Computer Science, teamed up to develop a highly effective, automated approach to diagnosing COVID-19. Also involved were PhD student Jenita Manokaran and research assistant Dr. Fatemeh Zabihollahy.

Ukwatta said the goal of this technology is “to alleviate the burden and to enable rapid COVID-19 diagnosis at the bedside.”

The research team trained a computer system to distinguish COVID-19 infections, pneumonia and normal lungs from chest X-ray images. This technology achieved 97-per-cent accuracy using artificial intelligence (AI), which instructs computers to complete tasks that would normally require human intelligence.

Dr. Andrew Hamilton-Wright

“AI-based technology can play a vital role in the diagnosis, follow-up and prognosis of the patients,” said Ukwatta.

By presenting the computer with images from patients worldwide, the researchers developed an algorithm that recognizes various lung conditions. This branch of AI is called “deep learning,” a type of machine learning that is best suited for identifying lung patterns in images.

“COVID-19-infected lungs manifest as a distinct pattern from other lung ailments. Deep learning methods [can] automatically extract high-level features from the images that distinguish between the normal and abnormal types,” said Ukwatta.

The machine learning algorithms can also be used to determine the severity of the COVID-19 diagnosis to help in allocating hospital resources.

“The technology helps to assess the severity of the disease affecting the lungs and potentially identify patients that may eventually require a ventilator for survival,” said Ukwatta. “Clinicians can make informed decisions to determine the need for hospitalization and effectively manage the patient population to provide best care.”

Efficient testing procedures are important in helping to defeat the pandemic, said Ukwatta. This diagnostic tool can minimize the burden on medical professionals and increase productivity in the health-care system.

This project was funded through U of G’s COVID-19 Research Development and Catalyst Fund and by the College of Engineering and Physical Sciences. For more information on how U of G researchers have tracked, mitigated, responded to and created through the pandemic, visit the Office of Research website.

Contact:
Dr. Eranga Ukwatta
eukwatta@uoguelph.ca