Quantitative evaluation of blood cells using a neural network on a microfluidic chip

Portable medical devices are becoming more common. The article considers the possibility of using an artificial neural network for counting blood cells on a microfluidic device. The proposed system is based on recording the absorption and scattering of light, when cells pass through the channel. The neural network was trained on samples containing only one type of cells. The ANS study was conducted on whole blood and diluted samples. The results obtained show that the number of errors admitted in the classification of leukocytes is much greater than necessary for practical use. Additional sample preparation significantly reduces the number of errors in the count of white blood cells. The classification of erythrocytes and platelets with ANS was satisfactory.