Bioimage analysis for multiplexed FUCCI acquisitions powered by deep learning

Abstract

The FUCCI sensor fluorescently labels cell cycle phases, which is essential to assess normal and abnormal cell-cycle progression in physiological and pathological conditions of developing organisms. However, accurate cell-cycle decoding is challenging in the low signal-to-noise conditions typical of multiplexed live cell imaging. To address this challenge, we developed deep learning networks that integrate FUCCI signals with a cytoplasmic alpha-tubulin fluorescent reporter. Our approach outperforms existing methods for both segmenting and classifying FUCCI nuclei, even in low signal-to-noise conditions. The resulting high-accuracy segmentation enables robust automated tracking. We leverage this to introduce a dynamic time warping analysis that determines cell cycle pseudotime from incomplete tracks and can detect cell cycle arrest. We provide pre-trained networks for multichannel FUCCI analysis, offering a powerful tool for studies in cancer research, development, and mechanobiology.