A neural network was taught to diagnose coronavirus from medical images with an accuracy of 80% at Innopolis University

Employees of the Center for Artificial Intelligence of a Russian IT university located on the territory of the Innopolis SEZ trained a neural network to diagnose coronavirus from medical images. Diagnostic accuracy is 80%, which will help doctors quickly identify patients with developing coronavirus pneumonia in mass diagnostics around the world.

The training took place using 28 thousand medical images with pneumonia, including 94 images taken from patients with COVID-19. Employees of the center are preparing to launch an online service that will help doctors around the world to identify patients with advanced coronavirus pneumonia.

Diagnosis of patients with suspected new coronavirus infection is multifactorial: patients undergo a blood test, a smear on the microflora of the nose and pharynx, but in a number of countries there are not enough tests, so only a small number of patients with severe symptoms or confirmed contact with patients are tested. In this case, an X-ray of the lungs becomes one of the available options for mass diagnosis: in some cases, special signs are observed in the images, which can only be caused by coronavirus pneumonia.

The algorithm and radiologist independently analyzed the images of patients with coronavirus. It turned out that the prediction of the model about the presence or absence of pathology coincides with the description of the doctor in 80% of cases. In addition, the algorithm was not mistaken in 13%, in which the doctor could not determine the pathology. It is expected that in the near future the sizes of datasets of X-ray images with coronavirus will increase significantly, which will significantly improve the accuracy of the algorithms.

Let us recall that the Ministry of Economy of Tatarstan is responsible for the development of one hundred industrial infrastructure facilities. In addition, the Ministry of Economy is the single regional coordinator for ensuring the development of the Innopolis SEZ and the Alabuga SEZ. In this status, the department is responsible for the development and transfer of high-tech solutions, improving the conditions for creating highly competitive assets in the field of intellectual property.