Multi-Object Tracking of Bone Osteosarcoma Epithelial Cells

When studying the response of cells to drugs and estimating proliferation status, bulk population measurements and single-time snapshots are being used. They are useful to get a global idea of gene expression differences between samples, but they are population- and time-averaged measurements. Therefore, they are an indirect measure of responses to chemotherapeutic drugs. Instead, live long-term high-temporal resolution assays can be done. They are more accurate in capturing the proliferation behavior than the standard approaches. To carry out such a study, the manual analysis of an expert is required. This involves analyzing when a cell dies or divides, and into which cells it divides. All of this has to be documented, which is not only very time-consuming but also costly.

In this project, we use deep learning methods for automatic tracking of bone cancer cells in microscopic videos. We investigate different approaches to detect the cells of the respective frames in the first step to match them across several frames in the second step. The challenge within the project is not only to track individual cancer cells but also to recognize events such as cell division and cell death. Such an automatic analysis allows to make statements about the proliferation behavior of the cells in response to drugs.