Morphology And Status Transition Monitoring For Mouse Embryonic Stem Cell Colonies In Virto By LSTM Networks With Progressive Training Using Fluorescence Microscopy Images

Abstract

Purpose Embryonic stem (ES) cells represent as a cellular resource for basic biological studies and for their uses as medically relevant cells in in vitro studies. Fluorescence microscopy images taken during cell culture are frequently used to manually monitor time-series morphology changes and status transitions of ES cell (ESC) colonies, and to study dynamical pattern formation and heterogeneity distribution within ESC colonies, intrinsic fluctuation and cell-cell cooperativity. Therefore, tracking and furthermore predicting morphology changes and status transitions of ESC colonies is an effective method to monitor culture medium for maintaining ES cells in undifferentiated or early differentiated stage. Methods A P-LSTM (Progressive Long Short-Term Memory) structure is proposed to incorporate some new time-lapse images real-time taken from incubators for a new RNN (Recurrent Neural Networks) training. The P-LSTM can achieve adaptive long- and short- term memories to generate accurate predicted images. On the time-lapse images, entropy and bi-lateral filtering are used to extract the range of every colony to calculate colony morphology. Colony status transitions between consecutive images are calculated by mapping the calculated colony centers and ranges. Results Accuracies for the colony status transition, area and roundness for the 15 predicted (five-hour) future frames calculated from 1500-2500 colonies for respective frames show the effectiveness of the proposed method.Conclusion We proposed an efficient and automatic method to predict and monitor status transitions and morphology changes of mouse ESC colonies in culture using time-lapse fluorescence microscopy images.