Centrala begrepp
Deep learning models accurately identify and analyze cell divisions in epithelial tissues, revealing insights into tissue growth and repair dynamics.
Sammanfattning
Cell division is crucial for tissue growth and repair responses. Deep learning models can automate the analysis of cell divisions in complex tissues, providing valuable insights into spatial and temporal patterns. The study developed deep learning models to detect dividing cells, determine their orientation, and analyze post-division behaviors. These models were trained on time-lapse imaging data of Drosophila pupal wing epithelium. The results show that cell divisions are influenced by tissue tension lines and exhibit spatial synchronicity following wounding. The study also highlights the importance of multiple fluorescent channels for accurate model performance. Overall, the deep learning approach offers a powerful tool for studying dynamic cell behaviors in living tissues.
Statistik
Accurate detection achieved 78.7% of all cell divisions.
U-NetCellDivision10 reduced false positives and negatives by over 80%.
Daughter cells shuffled post-division with a mean shift of 14.8°.
Post-shuffling daughter cells aligned along the global tension axis.
Cell divisions showed no bias towards wounds but exhibited spatial synchronicity post-wounding.