Tracking leukocytes in intravital time lapse images using 3D cell association learning network

Abstract

Leukocytes are key cellular elements of the innate immune system in all vertebrates, which play a crucial role in defending organisms against invading pathogens. Tracking these highly migratory and amorphous cells in in vivo models such as zebrafish embryos is a challenging task in cellular immunology. As temporal and special analysis of these imaging datasets by a human operator is quite laborious, developing an automated cell tracking method is highly in demand. Despite the remarkable advances in cell detection, this field still lacks powerful algorithms to accurately associate the detected cell across time frames. The cell association challenge is mostly related to the amorphous nature of cells, and their complicated motion profile through their migratory paths. To tackle the cell association challenge, we proposed a novel deep-learning-based object linkage method. For this aim, we trained the 3D cell association learning network (3D-CALN) with enough manually labelled paired 3D images of single fluorescent zebrafish's neutrophils from two consecutive frames. Our experiment results prove that deep learning is significantly applicable in cell linkage and particularly for tracking highly mobile and amorphous leukocytes. A comparison of our tracking accuracy with other available tracking algorithms shows that our approach performs well in relation to addressing cell tracking problems.