An Automatic Tracking Method for Multiple Cells Based on Multi-Feature Fusion

Cells automatic tracking in microscopy image sequences is an important task in many biomedical applications, especially for the analysis of anticancer drugs. However, it is still a challenging problem due to the high density, variable shape, lack of effective feature information, and occlusion of the cells by division or fusion. In this paper, the aim is to develop a fully automatic and effective method to track hundreds of cells, and a multi-feature fusion re-tracking algorithm is proposed based on the tracking-by-detection scheme. First, a region proposal method based on faster R-CNN is presented to generate cell candidate proposals. Then, a cell tracking method is proposed by fusing the bounding box and feature vector of cell candidates based on the above mentioned results. Finally, a re-tracking algorithm is employed by integrating historical information of matching frame. A series of experiments is conducted to test and verify the validity on the datasets from ISBI Cell Tracking Challenge, and then, the proposed method is applied to the T24 dataset of bladder cancer cells from the Cancer Cell Institute, University of Cambridge. The experimental results are encouraging and show that the proposed method is competitive with other state-of-the-art methods, which means that there are probably potential applications in the field of biomedical engineering.