Cell migration is a complex and heterogeneous phenomenon consisting of many spatially and temporally regulated mechanisms. A complete understanding of cell migration phenomenon is indeed crucial to derive an adequate quantification for single cell-based analysis. However, the spatiotemporal behavior of dynamic cells is often complex and challenging for manual classification and analysis. Automatized algorithms are thus highly desired to facilitate mesenchymal migration sub-modalities prediction. Supervised deep learning techniques have shown promising outcomes in microscopy image analysis. However, their implication is confided by the amount of carefully annotated data. Weak supervision provides a simple, model-agnostic way to integrate the domain-expertise into a learning model. Additionally, bayesian predicitons can lead to a more informed decision, and the quality of prediction can be improved. Currently, this study employs a Bayesian CNN regression model for predicting (or classifying) the probability of each cell observation.
High-resolution confocal images of a two example cell during migration in the Discontinuous and Continuous mode
when treated with 10 mg/ml fibronectin concentration.
Cell Predition pipeline.
Mesenchymal migration prediction visual results.
mesenchymal migration prediction plot.