A Preliminary Study on Retro-reconstruction of Cell Fission Dynamic Process using Convolutional LSTM Neural Networks

Abstract

Cell morphological analysis has great impact towards our understanding of cell biology. It is however technically challenging to acquire the complete process of cell cycles under microscope inspection. Using convolutional long short-term memory (ConvLSTM) networks, this paper proposes a comprehensive visualization method for cell cycles by retro-reconstruction of the preceding frames that are not captured. Results suggested that this method has the potential to overcome existing technical bottlenecks in image acquisition of cellular process and hence facilitate cell analysis.Clinical Relevance— This model allows back-tracing to complete the visualization of the cellular processes through a short segment of microscope-acquired cellular changes hence providing a starting point for exploring applications in predicting or backtracking unknown cellular processes.

Publication
2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)