Big data of big tissues: deep neural networks to accelerate analysis of collective cell behaviors in large populations

Coordinated cellular motion is crucial for proper tissue organization and function. Biophysical statistics of group behaviors can provide key insights into these behaviors, such as detecting pathological changes. Applying statistical analyses to very large tissues (>50,000 cells) offers exciting potential but requires more versatile feature extraction approaches. Convolutional neural networks are increasingly promising for tasks such as object classification and segmentation (e.g. cells, phenotypes). In this study, we apply a U-Net style architecture for label-free nuclei detection using low-magnification, transmitted light imaging. We utilize UV-excited nuclear labels to achieve automatic annotation of the data. The benefits of label-free image segmentation with transmitted light microscopy are numerous: improved accessibility by allowing for feature extraction without fluorescence imaging; eliminating the phototoxicity that results from typical UV-excited nuclear labels; and unambiguous, rapid post-processing of massive datasets. Here, we assess the accuracy of the reconstructed nuclei, and present preliminary data using this tool to explore cellular distribution and migration statistics (e.g. order, neighbor arrangements, correlations) in large, complex tissue geometries.