Automatic Counting System of Red Blood Cells Based on Fourier Ptychographic Microscopy

Red blood cell (RBC) counting is of great medical significance in clinical examination. Commonly, the cell counting task is completed by microscopic examination, which requires a high resolution. This paper proposes an automatic counting system of red blood cells based on Fourier ptychographic microscopy (FPM) and estimates the RBC number via a convolutional neural network (CNN). The counting network is based on a regression model, using a VGG-16 network combined with a feature pyramid network (FPN). The experimental results show that the mean absolute percentage error (MAPE) of our counting network can reach 0.86%, which means a high accuracy.