A deep learning approach to assess transendothelial cell trafficking performance

Transendothelial migration (TEM), is a complex, multistep process impacted by diseases like autoimmune disorders and cancer. Deciphering aspects of the process enhances our understanding and possibilities for disease treatments. The extent to which this potential can be leveraged is often limited due to conventional assays neither accurately mimicking specific in vivo conditions like sheer stress nor visualizing the whole transmigration cascade, hence missing distinct mechanisms. Flow-based adhesion assays overcome these limitations and allow the use of various endothelial cells from different tissues with distinct properties. So far, a broader and translational assay application is hampered by potential operator-based bias/lack of standardization, as well as poor scalability due to time-consuming manual analysis. In this study, we successfully combined this assay with AI-based analysis including subsequent classification of cell-transmigration phases by a Keras/TensorFlow-based deep learning model. Trained on healthy-donor and pancreatic cancer patient-derived T cells, the model achieved a high accuracy of 91.6 % in identifying/categorizing cell transmigration, surpassing the currently accepted 80 %-threshold, therefore qualifying as a fast, standardized AI-based live-cell imaging tool. Additionally, its architecture grants highly convenient reconfiguration for various disease-model investigations. Hence, by combining an affordable and simplistic, yet potent live-cell-imaging technique with a comprehensive AI-approach, we have established a powerful tool which allows for integrating TEM-assays into various disease models.