A research team led by Professor Yong-Keun Park and Professor Chan Hyuk Kim from the Department of Physics conducted a research concerning the application of a deep learning framework and 3D holographic microscopy to help speed up key processes of cancer research. The research was published on December 17, 2020 in the journal eLife under the title “Deep-learning-based three-dimensional label-free tracking and analysis of immunological synapses of CAR-T cells”.

When researchers use conventional microscopy techniques to study cell-to-cell interactions, they often damage the cells they study. Not only that, the amount of work scientists have to do to analyze these interactions is massive. However, the application of deep learning and 3D holographic microscopy can easily reduce the intensive workload and avoid cellular damage. For the first time, these techniques will be applied on the study of the immunological synapse.

A critical path to the understanding and development of cancer immunotherapy is the study of the formation of the immunological synapse (IS) junction. The current method of microscopy — fluorescence-based imaging — damages the cell when it is exposed to too much illumination, and therefore prevents the study of the changes in the IS junction. 3D holographic microscopy circumvents this problem by not being reliant on illumination; rather, it depends on the refractive index of different parts of the cell to create a 3D hologram. Despite its advantages, holographic microscopy has rarely been used for cell-to-cell interaction because it struggles to distinguish the parts of the different cells. Although humans can correct this mistake by manual segmentation, this is usually too difficult or time-consuming.

To overcome the challenges of manual segmentation, the research team applied a deep-learning framework, called DeepIS, on the segmentation problem. The framework is able to easily label and distinguish different parts of the IS junction formation process. To validate their research, the team allowed DeepIS to analyze the IS junctions formed between chimeric antigen receptor T-cells and target cancer cells. They then performed manual segmentation and compared the results. Astoundingly, DeepIS was able to capture information that even human manual segmentation can easily miss. Professor Yong-Keun Park, head of the research team, commented, “In addition to allowing us to avoid the drudgery of manual segmentation and the problems of photo-bleaching and photo-toxicity, we found that the AI actually did a better job.”