Herein, a calibration procedure to determine the depth positions of particles in a microfluidic channel via astigmatism particle tracking velocimetry (APTV) has been described. A neural network model focusing on the geometrical parameters of distorted particle images was developed to calibrate APTV. To demonstrate the efficiency of this procedure, the Poiseuille flow and depth of the particles, and dispersions in the microchannel were studied. The depth positions were determined with an uncertainty of ±1µm. The present results suggest that the particle position dispersion could be a result of the degree of particle image deformation and its deviation.