Researchers have developed an AI tool to detect COVID-19 cases from images.

Linda Wang and Alexander Wong, researchers at the Waterloo Artificial Intelligence Institute, Canada, have developed COVID-Net. It is a Deep Convolutional Neural Network designed to detect COVID-19 cases from chest X-rays. (AI in Healthcare)

For a deeper understanding, and further development of the AI network, the researchers have open-sourced and made it available to the general public. They are also making available COVIDx, the chest radiography dataset. They trained  COVID-Net on this collection of 5941 chest x-ray images from 2839 patient cases.

Covid-19 and AI

Rapid, effective, and accurate screening of patients to detect the coronavirus is extremely important to fight its spread and deadly payload. Healthcare providers must immediately isolate infected patients and provided them with immediate treatment and care.

The current and accurate screening method is polymerase chain reaction (PCR) testing on respiratory specimens from nasal or ear swabs. Though it is highly sensitive and accurate, PCR is a very time-consuming and laborious process.  It is a complicated manual process and already in extremely short supply.

However, doctors can achieve a reasonably dependable diagnosis through the examination of patients’ chest x-rays. The radiographic images of confirmed Covid 19 patients display marked abnormalities, however. The quick identification of such characteristic abnormalities in the x-rays of patients in epidemic areas could be a far more effective means to identify and isolate those that are infected.

A computer-aided diagnostic system using AI could scan these x-ray images more efficiently compared to human evaluation.

COVID-Net, the open-source, deep convolutional neural network, is such a tool. The need for faster interpretation of radiographic images for the detection and treatment of the deadly COVID-19 disease was the motivation for its development at the Waterloo Artificial Intelligence Institute.

Analysis of results

The researchers performed a quantitative analysis to determine the accuracy of the system. “It can be observed that COVID-Net strikes a good balance between accuracy and computational complexity by achieving 83.5% test accuracy while requiring just 2.26 billion MAC operations to perform case prediction.”

However, the researchers observed that though the system performs well in its objective of detecting Covid 19 infections from chest x-rays, there was room for improvement. The collection of additional data, as well as enhancement of the underlying training methodology, could generate better results.

The researchers take care to point out that COVID-Net is not “a production-ready solution.” However, they hope that its promising results could lead to further development by both researchers and citizen data scientists.

That would accelerate treatment for those who need it most.

Another tactic: detection by CT scans

Alibaba’s (NYSE: BABA) research institute Damo Academy developed an AI-based disease algorithm. The company deployed the CT-scan based system at the new Qiboshan Hospital in Zhengzhou, Henan province. The algo is able to complete the recognition process for coronavirus in a patient within 20 seconds with 96% accuracy.

Because it is able to speed up the recognition process, and therefore the treatment action for the disease, the algo is now being adopted in over 100 hospitals in the provinces of  Hubei, Guangdong, and Anhui.

Related Story:    Artificial Intelligence: Alibaba’s AI System Detects Coronavirus With 96% Accuracy