Docs come up with artificial intelligence system to identify COVID-19 with X-rays
Scientists have developed an online system for detecting early signs of coronavirus using X-rays.
By analysing a patient’s chest X-rays with their artificial intelligence CIRCA system, the scientists from the Silesian University of Technology (SUT) are able to distinguish coronavirus symptoms from other pulmonary infections.
This will help support and accelerate COVID-19 imaging diagnostics in hospital emergency departments, facilitating the initial assessment of the nature of changes in the lungs of patients with respiratory disorders.
Professor Joanna Polańska, head of the Faculty of Automatic Control, Electronics and Computer Science at SUT, who led the research team said: "Because disease symptoms in the early stages of pneumonia and COVID-19 are mostly very similar, there was a need to create a tool for non-radiologists that could reassure them that the image acquired in the X-ray examination indicates the development of the disease towards COVID-19.”
The researchers used artificial intelligence (or machine learning) to develop a programme capable of distinguishing images showing three basic lung conditions: image without lesions (healthy lungs), with changes characteristic of pneumonia in COVID-19 and with changes characteristic of other types of inflammation (bacterial and viral).
Polańska said: “By analysing every smallest detail of an X-ray image CIRCA is able to extract the characteristic features, the so-called “radial signature” of each disease entity, including COVID-19.
“The proposed radiogram analysis system is a combination of several convolutional and classic neural networks, which together with a number of supportive procedures allow for automatic identification of chest x-rays, segmentation of the lung region and identification of cases visually changed as a result of the development of COVID-19, “ she added.
As part of the scheme, a nationwide PolCOVID image database will be created to allow for retrospective in-depth analysis of tens of thousands of X-ray images. It will help develop detailed diagnostic guidelines for clinicians and identification of the COVID-19 radial signature at various stages of the disease advancement.
In the future, the detailed analysis of data collected during the epidemic, supported by the opinions of radiologists will make it possible to determine the radiometric features and their variability characteristic of COVID-19.