Event Detection by Feature Unpredictability in Phase-Contrast Videos of Cell Cultures

摘要: In this work we propose a novel framework for generic event monitoring in live cell culture videos, built on the assumption that unpredictable observations should correspond to biological events. We use small set of event-free data train multioutput multikernel Gaussian process model operates as an predictor by performing autoregression bank heterogeneous features extracted from consecutive frames video sequence. show prediction error can be used probability measure presence relevant events, enable users perform further analysis or large-scale non-annotated data. validate our approach two phase-contrast sequence sets containing mitosis and apoptosis events: new private dataset human bone cancer (osteosarcoma) cells benchmark stem cells.