Innopolis university: Service of diagnosis of koronavirusny pneumonia

2020: Preparation for start of service of diagnosis of koronavirusny pneumonia

On March 26, 2020 the Innopolis University reported that the staff of the Center of artificial intelligence Russian trained a neuronet at 28 thousand medical images with pneumonia, including at 94 images taken from sick COVID-19. The staff of the center prepares for start an online service which will help doctors to identify quicker patients with the developed koronavirusny pneumonia at mass diagnostics around the world.

Diagnosing of patients with suspicion of a koronavirusny infection multifactorial: do to sick blood test, dab on a microfloor of a nose and pharynx, but in a number of the countries there are not enough tests therefore analyses do only to a small number of patients with heavy symptoms or the confirmed contacts with patients. X-ray of lungs in that case becomes one of available options of mass diagnostics: in certain cases in pictures special signs which koronavirusny pneumonia can cause are observed.

The command of the Center of artificial intelligence of the University Innopolis since 2014 works on algorithm elaboration and services of recognition of medical images, including x-ray images of bodies of a chest cavity, the KT-image and the MRT-image of different bodies, on the basis of neural networks. One of developments — an algorithm of detection of symptoms of pneumonia according to x-ray images. For training of artificial intelligence 28 thousand pictures of easy healthy people and patients with different types of pneumonia included in a data set. Specialists of IT university adapted model for the arisen task and turned on from open sources in a data set of an algorithm of 94 x-ray images of bodies of a chest cavity of the infected COVID-19 taken from an open data set on the website GitHub which is replenished every day. So, the neural network learned to define the general symptoms of the pathologies caused by a coronavirus.

As the head of the Center of artificial intelligence of the University explains Innopolis Ramil Kuleev, the accuracy of a neuronet is still insufficient for povmestestny clinical implementation, however the method after completion will be useful at mass researches of lungs:

Annually in Russia about 80 million fluorographic researches are conducted, and in day it is about 220 thousand pictures. And, presumably, epidemiological situation changed the number of procedures little. Tests for a virus much less, and they require additional resources therefore mass screening and the analysis of pictures of bodies of a chest cavity will help to identify the patients with a coronavirus among the population and also to plan treatment. Besides, the service can be deployed in a cloud that will allow to connect X-ray rooms of the hardly accessible and remote settlements to our system. Automatic data analysis will increase efficiency of diagnostics and will lower load of radiologists,

Algorithm and the radiologist independently analyzed pictures of patients with a coronavirus. It turned out that prediction of model for existence or absence of pathology matches the description of the doctor in 80% cases. Besides, the algorithm was not mistaken in 13% in which the doctor could not define pathology. It is expected that in the nearest future the sizes of data sets of x-ray images with a coronavirus will significantly increase. It will allow to increase the accuracy of algorithms considerably.

When the doctor-clinical physician faces the patient with symptoms of a koronavirusny infection, in the arsenal he has very limited tool kit, allowing to reveal quickly symptoms of a disease. The radiological research of bodies of a chest cavity is the objective tool for identification of changes in lungs which can be signs of development of koronavirusny pneumonia. The result of digital roentgenography which is quickly received by the doctor could help with identification at patients with a coronavirus of the progressing changes in lungs. The software for automatic recognition of symptoms of a disease according to x-ray images could simplify work of the doctor-clinical physician and increase efficiency of diagnostics in need of carrying out urgent researches, including at impossibility of the instant analysis of a research the radiologist. But at the same time, of course, questions on embedding of a system evolve from artificial intelligence in an algorithm of diagnosis of a koronavirusny infection for a possibility of the most operational identification of the developing pathological changes in lungs and acceptances of urgent measures for effective treatment, noted the manager of X-ray diagnostic department of the Republican clinical antitubercular clinic of Tatarstan Sergey Konovalov

Lack of laboratory tests for COVID-19 — one of the main problems of early diagnosis of a disease for a timely initiation of treatment of the patient and minimization of his contacts with people around. The Radiology magazine published a research of KT-pictures of passengers from the Diamond Princess ship. More than a half of passengers who had no disease symptoms yet had shading in pulmonary fields. Fluorographic researches have a diagnostic potential also — cheaper, safe and widespread tool for the analysis of bodies of a thorax. Our command is engaged in development of methods of artificial intelligence for diagnosis of diseases of lungs and could already develop algorithms for diagnosis of pneumonia at the level, close to doctors. I will note that the analysis of pictures will not replace use of laboratory tests, but can seriously help where access to tests is complicated, commented the Assistant-professor to the University of Copenhagen and the leading researcher of the University of Innopolis Bulat Ibragimov