AI analyzes video of blood flow in capillaries, made using capillaroscopy. The accuracy of determining the content of leukocytes is 90%. With the help of the new method, it will be possible to find out the composition of the blood faster, without waiting for laboratory results.

Alexey Kornaev, research fellow at the Research Center for Artificial Intelligence at Innopolis University: "As a result of our research, we have learned to identify larger white blood cells - leukocytes - with an accuracy of over 90%. The proposed method with training on randomly selected fragments of the image increased the accuracy of counting blood cells by about 10%. Experiments also showed that training models on video is faster, and predictions are more accurate compared to training on static images. Of course, our method provides information only about high or low content of red blood cells and leukocytes in the blood, but even in this formulation it is of practical interest for medicine. We plan to link further research with the collection and processing of real video images of capillaroscopy.”

A detailed description of the experiment is published in the proceedings of the scientific conference International Conference on Learning Representations (ICLR).