A unique neural network has been created in Russia to determine the type of breathing

MOSCOW, November 15. An AI model for automatically determining the type of breathing during lung therapy was created by specialists from Volgograd State Technical University. According to them, the development, which has no analogues, can work at home with a conventional camera, which expands the possibilities of rehabilitation. The results were published in the journal Algorithms.
According to scientists, special breathing training is one of the most important elements of rehabilitation after a number of diseases; they can also be an effective way to reduce stress and generally strengthen the body.
For this purpose, technical systems are being actively developed all over the world that are capable of monitoring the quality of training with an accuracy that exceeds the capabilities of a human trainer.

The team of the Volgograd Innovation Laboratory for Data Analysis and Management (V.I.S.D.O.M.) of the Volgograd State Technical University (Volgograd State Technical University) of the Department of Automated Systems Software, together with colleagues, created a model for the rehabilitation of people with pathologies of the respiratory system. According to the authors, the development makes it possible to organize effective training even at home.

«

“To train the computer to determine the type of breathing, which is the starting point in the training program, we had to independently generate an extensive data set, recording how a person breathes with abdominal, thoracic or mixed breathing. There are no similar systems in the world today,” said Yulia Orlova, head of the department of software of automated systems at Volgograd State Technical University.

Determination of the type of breathing can occur with equal efficiency using a motion capture system, which is more suitable for specialized exercise therapy — clinic rooms, and using data from a regular digital camera.

Experimental testing of the method, according to scientists, showed that in more than 80% of cases the neural network coped with the task correctly.
» «We wanted to create a universal tool that can help a rehabilitation trainer or even replace him. In the future, our AI model can also be trained to control exercises for restoring the musculoskeletal system,» said Orlova.
The research was carried out with partial support from program «Priority 2030».