In biological radiation research, cellular responses after irradiation are usually studied in vitro, i.e. on cells in so-called cell culture. The most important endpoints are proliferation (number of divisions per cell), cell cycle progression and cell death. This information can then be used to infer what happens in irradiated tissue in the body, which is important, for example, in order to be able to use radiation ideally in radiotherapy or to be able to protect the body better from radiation. Radiation biology endpoints are usually evaluated using standardized procedures such as the colony formation assay for proliferation or the caspase 3/7 sytox assay for cell death. However, the information gain of these procedures is severely limited. For example, these procedures can only be evaluated at a specific time point and only a few endpoints can be looked at depending on the procedure. Thus, one does not get any information about the temporal development of the endpoint under investigation and also no information about how different endpoints are related. Unfortunately, the standard methods are also very time-, cost- and staff-intensive and therefore cannot be repeated as often as desired. In addition, these methods require special treatments of the cells with, among other things, chemicals that affect the reactions in the cell.

To solve these problems, we are developing a new method here at the institute: We microscope the cells after irradiation with phase contrast and analyze the resulting videos with the algorithm CeCILE (Cell Classification and In-vitro Lifecycle Evaluation), which is based on artificial intelligence. CeCILE is designed to detect each individual cell and track it across the entire video. From this, the proliferation, i.e. how often each cell has divided, the course of each cell cycle and the time and circumstance of each cell death can then be evaluated. Thus, not only the temporal component can be added with our method, but the interrelationships of the individual endpoints can now be explored. Microscopy with phase contrast also has the advantage that no dyes or other treatments are necessary, allowing the cells to be observed completely undisturbed for several days. In the publication below, we have already shown that CeCILE provides comparable results to standard methods but can provide more information about the temporal component. For the development of CeCILE, an own data set was created on which the artificial intelligence was trained. The dataset itself is continuously being improved to make the training optimal. We are also currently working on improving the object detector and implementing a tracking method.

Students interested in  Deep Learning and artificial intelligence in image processing can contact Prof. Judith Reindl or Sarah Rudigkeit for a student research, bachelor or master thesis. Currently we are looking for a motivated master student for a master thesis, here.