Scientists from the University of Waterloo, in collaboration with a Canadian Artificial Intelligence firm has developed an open-access neural network that can speed up diagnosis of COVID-19 infections.
Developers Linda Wang and Alexander Wong partnered with DarwinAIto come up with a convolutional neural network named COVID-Net. Such networks specialize in image recognition through algorithms. The current solution was trained using 5,941 chest x-ray images from 2,839 patients with different lung ailments. The data set included x-rays of COVID-19, as well as other viral and bacterial infections. Based on the inputs, researchers claim that COVID-Net could identify signs of COVID-19 infections from chest x-rays with a fair amount of success.
Amidst the recent pandemic scenario, there has been several claims of diagnostic AI tools for COVID-19 that purportedly use x-ray imaging. However, none of them has been amply backed with supporting evidence.
In this context, DarwinAI has taken a surprisingly open approach. They have candidly announced that it is “by no means a production-ready solution” and have opened up access to the COVID-Net tool along with the entire test data set. The idea is anyone can explore and refine it further if possible.
If you really want to have a go at it, the source code, documentation, dataset, and the relevant scientific paper are all available at the following GitHub repository: