Two-photon intravital microscopy (2P-IVM) has become a widely used technique to study cell-to-cell interactions in living organisms. Four-dimensional imaging data obtained via 2P-IVM are classically analyzed by performing automated cell tracking, a procedure that computes the trajectories followed by each cell. However, technical artifacts, such as brightness shifts, the presence of autofluorescent objects, and channel crosstalking, affect the specificity of imaging channels for the cells of interest, thus hampering cell detection. Recently, machine learning has been applied to overcome a variety of obstacles in biomedical imaging. However, existing methods are not tailored for the specific problems of intravital imaging of immune cells. Moreover, results are highly dependent on the quality of the annotations provided by the user. In this study, we developed CANCOL, a tool that facilitates the application of machine learning for automated tracking of immune cells in 2P-IVM. CANCOL guides the user during the annotation of specific objects that are problematic for cell tracking when not properly annotated. Then, it computes a virtual colocalization channel that is specific for the cells of interest. We validated the use of CANCOL on challenging 2P-IVM videos from murine organs, obtaining a significant improvement in the accuracy of automated tracking while reducing the time required for manual track curation.