Reception of the district executive committee

Address:
4 Engelsa Street
247960, Zhitkovichi

E-mail: isp@zhitkovichi.gov.by

11
Main News Republic

Republic

9 April 2021

Belarusian neural network detects asymptomatic pneumonia

MINSK, 9 April (BelTA) – The Belarusian startup Covid-Ray and the cloud provider IBA Cloud have launched a service that can detect even asymptomatic pneumonia with 91% accuracy by processing a digital image, BelTA has learned.
The service can detect pulmonary fever by processing an x-ray image. The artificial intelligence determines the patient's status – normal, pneumonia, another pathology – on the basis of about 250,000 images. Every diagnosis has a degree of accuracy and it is up to human medics to make the final diagnosis. The latest experiment demonstrated a degree of accuracy even higher than that of highly specialized doctors.
A neural network is at the heart of the service. It is taught using virtual servers with GPUs and fast NVMe drives. The developers even shifted the software from Amazon to Belarus since it turned out to be more effective and economical.
Belarusian healthcare institutions already use the service. They upload hundreds of images of patients, who may or may not have pulmonary fever. If the neural network finds one, specialized doctors take a closer look at the image. The approach has improved the quality of diagnostics of Belarusians, whose pneumonia shows weak symptoms or no symptoms, including during a coronavirus infection.
Foreign clinics also use the capabilities of the Belarusian startup. The geography is impressive – from London to Vladivostok. Medics from New Zealand also took a look at the platform the other day.
The project's creator Andrei Kapitonov said: “The service is particularly interesting now, during a pandemic because it allows processing x-ray images and detecting lung pathologies promptly and with a high degree of precision.”
The software cannot replace medics but it can improve diagnostics precision. At present a human radiologist in a clinic examines 100-150 x-ray images per day and the heavy workload predictably reduces the quality of the work. Unlike people neural networks do not tire and process every image with a high degree of precision. It is particularly important that the service produces virtually no false negatives. If something is wrong with the patient's lungs, it will be noticed, the Covid-Ray developers assured.

Belarus news (BELTA)