Neural network was trained to identify coronavirus by images
Scientists have developed a neural network to udentify coronavirus by X-rays and computed tomography (CT) images, TASS reports. It, according to department head of the Novosibirsk State University of Informatics and Management (NSUIM) Sergei Tereshchenko, was trained using ten thousand photographs.
Tereshchenko noted that the computer vision method was used to detect pneumonia caused by COVID-19. For this, an artificial neural network was created, including a set of algorithms that works on the principle of networks of nerve cells in living organisms.
Having been trained on a large dataset, the neural network can perform various tasks now; for example, recognize X-rays almost at the human level.
The specialist said that that data used to “train” the neural networks are in the public domain. Most of the data was collected from hospitals in Novosibirsk.
Doctors participated in training the neural network. They provided the developers with a feature map that identifies the coronavirus. After training, the neural network identified traces of COVID-19 with an accuracy of 91%.
Scientists have already made the service publicly available on the Internet. Anyone can find out about the presence of coronavirus using their picture. In the future, it is planned to train the neural network to detect other lung diseases.