Scientists from Volgograd State Medical University (VolgSMU), together with colleagues from the Iranian Islamic Azad University, have created a neural network for diagnosing respiratory diseases. The system distinguishes flu from ARVI and adenovirus infection in real time — even before obtaining laboratory test results, TASS was told by Ekaterina Belikova, Associate Professor of the Department of Infectious Diseases at VolgSMU.
The neural network analyzes several data streams at once: temperature, pressure, glucose level, characteristic symptoms, and photographs of laboratory forms. The built-in optical character recognition module itself reads the test results from the photographs — there is no need to enter the numbers manually. The architecture is based on an improved network with long short-term memory (LSTM).
The symptoms of influenza, colds, and adenovirus infection are so similar that even an experienced therapist often makes mistakes at the first appointment. Experiments have shown that the model outperforms competing methods in diagnostic accuracy. The system does not claim to completely replace the doctor — the final decision remains with the specialist, but he receives a probable diagnosis instantly.
